feat(providers): add Auriko as built-in provider plugin - #2
Open
zxyaction wants to merge 146 commits into
Open
Conversation
- memory_setup.py: use shlex.split() for plugin dep checks instead of shell=True - transcription_tools.py: avoid shell=True for auto-detected whisper commands (user-provided templates via env var still use shell=True for compatibility) - cli.py: add comment clarifying intentional shell=True for user quick_commands - Add test verifying auto-detected template is shlex-safe Addresses CONTRIBUTING.md Priority NousResearch#3 (Security hardening — shell injection).
zxyaction
force-pushed
the
feat/auriko-provider
branch
from
May 13, 2026 18:34
08b125a to
070ba58
Compare
zxyaction
changed the title
zxyaction
force-pushed
the
feat/auriko-provider
branch
3 times, most recently
from
May 13, 2026 19:26
94e5b59 to
f41b3e2
Compare
Adds an explicit API compatibility mode prompt to the `hermes model -> custom`
flow so Codex-compatible third-party endpoints (and any other non-default
backend whose URL doesn't match the existing heuristics in
`_detect_api_mode_for_url`) can be selected explicitly instead of silently
falling back to chat_completions.
Choices: Auto-detect / chat_completions / codex_responses / anthropic_messages.
Persists `api_mode` to:
- `model.api_mode` (active session config)
- the matching `custom_providers[*]` entry (so re-activating the named
provider next time replays the same transport)
Salvaged from PR NousResearch#6125 onto current main: kept the new prompt and the
`_save_custom_provider(api_mode=...)` plumbing; the named-custom flow
already extracts and applies `api_mode` from the saved entry on current
main so those changes are preserved as-is. Test fixtures updated for the
new prompt and the existing display-name prompt.
Co-authored-by: littlewwwhite <1095245867@qq.com>
Removes the 'Launch hermes chat now? (Y/n)' prompt at the end of hermes setup. The summary already prints 'Ready to go! → hermes' so the auto-launch was redundant, and on macOS 26+ it could crash in prompt_toolkit when setup was invoked from the curl install script with stdin redirected from /dev/tty (NousResearch#5884, NousResearch#6128). After setup, users run 'hermes' themselves like every other CLI tool. Same pattern applies to the Windows installer. Closes NousResearch#6128 (narrower env-var-guarded fix superseded by removing the prompt outright).
skill_view ran the direct-path strategy across every skill dir before the recursive strategy, so a top-level skill in an external dir could silently shadow a same-named nested local skill. /skills correctly listed the local version (deduped local-first by _find_all_skills) but skill_view loaded the external one — confusing, and a real bug class for users with skills.external_dirs registered alongside categorized local skills. Pick a louder fix than @polkn's PR NousResearch#6136 proposed: collect every match across all dirs (direct path, recursive by parent dir name, legacy flat <name>.md), and if there's more than one, refuse with an error that surfaces every matching path plus a hint to load by the categorized form. Local-first precedence would have replaced silent external-shadowing with silent same-name collisions between two externals, or made an externally-shadowed-by-local skill unreachable by bare name with no signal. Refusing forces the user to disambiguate once and never wonder which skill ran. Recovery: pass the full categorized path ("foundations/runtime/explore-codebase" instead of "explore-codebase"), or rename one of the colliding skills. Co-authored-by: pol <pol.kuijken@gmail.com>
Recover from SIGWINCH without clearing the physical screen or scrollback buffer. The startup banner and tool summary are printed before prompt_toolkit owns the live chrome, so they live in normal terminal scrollback. Calling erase_screen() + \x1b[3J] on every resize removed that UI permanently — _replay_output_history cannot reconstruct it because the banner was never added to _OUTPUT_HISTORY. Instead, just reset prompt_toolkit's renderer cache and invalidate so the next incremental redraw starts from a clean slate, then let the original on_resize handler recalculate layout for the new terminal size. This matches the behaviour of bash/zsh/fish on SIGWINCH. Fixes NousResearch#22999
…rch#25071) * tui: make URLs clickable + hover-highlight in any terminal Problem ------- URLs printed by `hermes --tui` were not clickable in basic macOS Terminal.app. Cmd+click did nothing, the cursor didn't change shape — like nothing was detected — even though arrow buttons and other Box onClick handlers worked fine. Root cause ---------- Two layers of dead plumbing: 1. `<Link>` only emitted the underlying `<ink-link>` (which carries the hyperlink metadata into the screen buffer) when `supportsHyperlinks()` said yes. On Apple_Terminal that's false, so the per-cell hyperlink field stayed empty, so `Ink.getHyperlinkAt()` had nothing to return on click. The visible underline was just decorative. 2. `Ink.openHyperlink()` calls `this.onHyperlinkClick?.(url)`, but `onHyperlinkClick` was never assigned anywhere in the codebase. The click pipeline (`App.tsx → onOpenHyperlink → Ink.openHyperlink`) ran but bailed silently on the optional chain. Bonus discovery: even when wired up, there was no hover affordance — terminal apps can't change the system mouse cursor, so users had no visual signal that a cell was clickable. Arrow buttons in the chrome worked because they had explicit `<Box onClick>` styling; inline link URLs didn't. Fix --- - `Link.tsx`: always emit `<ink-link>` regardless of terminal capability. The renderer's `wrapWithOsc8Link` already gates the actual OSC 8 escape on `supportsHyperlinks()` further down — so terminals that don't understand OSC 8 still don't see the escape, but the screen-buffer metadata (which the click dispatcher reads) is now populated everywhere. - `ink.tsx + root.ts`: add `onHyperlinkClick?: (url: string) => void` to `Options` / `RenderOptions`, wire it to the existing `Ink.onHyperlinkClick` field in the constructor. - `src/lib/openExternalUrl.ts`: small platform-aware opener using `child_process.spawn` with arg-array (no shell) — http(s) only, rejects `file:`, `javascript:`, `data:`, etc., so a hostile model can't trigger arbitrary local handlers via `<Link url="file:///...">`. Detached + stdio ignore so closing the TUI doesn't kill the browser and Chrome stderr doesn't leak into the alt screen. - `entry.tsx`: pass `onHyperlinkClick: openExternalUrl` to `ink.render`. - `hyperlinkHover.ts` + Ink hover wiring: track the URL under the pointer in `Ink.hoveredHyperlink`, update it from `dispatchHover`, and inverse- highlight every cell of the matching link in the render-pass overlay (same pattern as `applySearchHighlight`). This is the cursor-hover affordance for clickable links — terminals don't expose cursor shape, so we light up the link itself. - `types/hermes-ink.d.ts`: add `onHyperlinkClick` to the `RenderOptions` shim so consumers (`entry.tsx`) type-check against the new option. Tests ----- - `src/lib/openExternalUrl.test.ts` (15 cases): http(s) accepted; file/js/ data/mailto/ftp/ssh rejected; macOS open(1), Windows cmd.exe start with empty title slot, Linux xdg-open dispatch; shell-metacharacter URLs pass through unmolested as a single argv element; synchronous spawn failure returns false. Verified empirically in Apple Terminal 455.1 (macOS 15.7.3): clicking a URL opens in default browser, hovering inverts the link cells, and moving away clears the highlight. Full TUI suite: 713 passing, 0 type errors. Reverts ------- The earlier attempt that version-gated Apple_Terminal in `supports-hyperlinks.ts` was based on a wrong assumption — Terminal.app silently strips OSC 8 sequences but does not render them as clickable hyperlinks. Reverted to the original allowlist. * tui: address Copilot review — explorer.exe on win32 + comment fixes - openExternalUrl: switch win32 from `cmd.exe /c start` to `explorer.exe`. cmd.exe's `start` builtin reparses the URL through cmd's tokenizer, so `&`, `|`, `^`, `<`, `>` either split the command or get reinterpreted — breaking both the protocol-allowlist safety story AND plain http(s) URLs with `&` in query strings. `explorer.exe <url>` invokes the registered protocol handler directly with no shell. - openExternalUrl.test.ts: rename the win32 test to reflect the new contract and add two regression tests — one with `&|^<>` metachars, one with the common analytics-URL `&` query-param pattern — both pinned to single-argv-element delivery via explorer.exe. - Link.tsx: fix misleading comment. OSC 8 escapes are emitted unconditionally by the renderer (`wrapWithOsc8Link` in render-node-to-output.ts, `oscLink` in log-update.ts). Non-supporting terminals silently strip the sequence, which is why hover/click affordance has to come from the in-process overlay rather than the terminal's own link rendering. Verified: 715/715 tests pass, type-check + build clean. * tui: address Copilot review #2 — async spawn errors + hover scope + docs 1. openExternalUrl: attach a no-op `'error'` listener on the spawned child BEFORE unref(). spawn() returns a ChildProcess synchronously even when the binary is missing (ENOENT on xdg-open / explorer.exe), unreachable, or otherwise unusable; the failure surfaces later as an 'error' event. An unhandled 'error' on an EventEmitter crashes Node, which would tear down the whole TUI. The listener is a deliberate no-op — we already returned `true` synchronously and the user just doesn't see the browser pop. 2. openExternalUrl.test.ts: add a regression test using a real EventEmitter to simulate the async-error path. Pins both the listener-attached contract and the "doesn't throw on emit" behavior. Was 17/17, now 18/18. 3. ink.tsx dispatchHover: bypass `getHyperlinkAt()` and read `cellAt(...).hyperlink` directly. `getHyperlinkAt` falls back to `findPlainTextUrlAt` for cells without an OSC 8 hyperlink, but the render-pass overlay (`applyHyperlinkHoverHighlight`) only matches on `cell.hyperlink === hoveredUrl` — so plain-text URLs would burn re-renders without ever producing the highlight. Hover is now a strictly 1:1 fit for what the overlay can paint. Plain-text URLs still get the click action via the existing dispatch path. 4. root.ts + ink.tsx doc comments: replace the misleading "typically `open` / `xdg-open` / `start` shell" wording with the actual safe recipe — argv-array spawn into `open` / `xdg-open` / `explorer.exe`, with an explicit warning that `cmd.exe /c start` reparses the URL through cmd's tokenizer and is unsafe + breaks `&`-query URLs. Verified: 716/716 tests pass, type-check + build clean. * tui: address Copilot review NousResearch#3 — hover damage, alt-screen cleanup, opener allowlist 1. ink.tsx onRender: stop folding steady-state hover into hlActive. hlActive forces a full-screen damage diff so previous-frame inverted cells get re-emitted when the highlight set changes. The transition IS the trigger — enter / leave / change-to-other-link. While the pointer just sits on a link the painted cells don't change and the per-cell diff handles the no-op. Folding the steady state in would burn a full-screen diff on every frame. Added a lastRenderedHoveredHyperlink tracker and gate the hlActive bump on `hovered !== lastRendered`. 2. ink.tsx setAltScreenActive: clear hoveredHyperlink (and the tracker) when toggling alt-screen state. Hover dispatch is alt-screen-gated, so once we leave there's no path to clear it. Without this, remounting <AlternateScreen> would paint a phantom hover from the previous session until the next mouse-move arrived. 3. openExternalUrl.ts openCommand: allowlist linux + the BSD family for xdg-open and return null for everything else (aix, sunos, cygwin, haiku, etc.). Previously the default-fallback always returned xdg-open, which made the caller's `if (!command) return false` dead and yielded a misleading `true` on platforms that probably don't have xdg-open. New tests cover the null path AND the openExternalUrl-returns-false-without-spawning behavior. Verified: 718/718 tests pass, type-check + build clean. * tui: address Copilot review NousResearch#4 — doc comment accuracy 1. openExternalUrl return-value doc: now lists all three false paths (URL rejected / no opener for platform / synchronous spawn throw) plus a note that async 'error' events still return true because the spawn was attempted. 2. ink.tsx onHyperlinkClick field doc: clarifies the callback receives either an OSC 8 hyperlink OR a plain-text URL detected by findPlainTextUrlAt — App.tsx routes both into the same callback. 3. hyperlinkHover applyHyperlinkHoverHighlight doc: drops the misleading 'caller forces full-frame damage' promise. Caller decides; for hover the current caller only forces full damage on transitions. No behavior change. 718/718 tests pass. * tui: address Copilot review NousResearch#5 — lint fixes 1. ink.tsx: reorder `./hyperlinkHover.js` import before `./screen.js` to satisfy perfectionist/sort-imports. 2. Link.tsx: drop unused `fallback` parameter destructuring + the trailing `void (null as ...)` dead-statement (would trip no-unused-expressions). Kept `fallback?: ReactNode` on the Props interface as a documented compat shim so existing call sites still compile, with a comment explaining why it's no longer wired up. 3. openExternalUrl.test.ts: replace `typeof import('node:child_process').spawn` inline annotations (forbidden by @typescript-eslint/consistent-type-imports) with a `SpawnLike` type alias backed by a real `import type { spawn as SpawnFn }`. No behavior change. 718/718 tests pass, type-check clean, lint clean on all modified files.
Keep the outer history_offset when _run_agent drains queued follow-ups recursively so transcript persistence includes every queued turn in the chain instead of only the last one.
Replace tenant-specific example text in the transcript offset regression with generic follow-up turns so the upstream test documents the bug without customer-specific wording.
…backends (NousResearch#25126) * feat(video_gen): unified video_generate tool with pluggable provider backends One core video_generate tool, every backend a plugin. Mirrors the image_gen + memory_provider + context_engine architecture: ABC, registry, plugin-context registration hook, and per-plugin model catalogs surfaced through hermes tools. Surface (one schema, every backend): - operation: generate / edit / extend - modalities: text-to-video (prompt only), image-to-video (prompt + image_url), video edit (prompt + video_url), video extend (video_url) - reference_image_urls, duration, aspect_ratio, resolution, negative_prompt, audio, seed, model override - Providers ignore unknown kwargs and declare what they support via VideoGenProvider.capabilities() — backend-specific quirks stay in the backend, the agent learns one tool Backends shipped: - plugins/video_gen/xai/ — Grok-Imagine, full generate/edit/extend + image-to-video + reference images (salvaged from PR NousResearch#10600 by @Jaaneek, reshaped into the plugin interface) - plugins/video_gen/fal/ — Veo 3.1 (t2v + i2v), Kling O3 i2v, Pixverse v6 i2v with model-aware payload building that drops keys a model doesn't declare Wiring: - agent/video_gen_provider.py — VideoGenProvider ABC, normalize_operation, success_response / error_response, save_b64_video / save_bytes_video, $HERMES_HOME/cache/videos/ - agent/video_gen_registry.py — thread-safe register/get/list + get_active_provider() reading video_gen.provider from config.yaml - hermes_cli/plugins.py — PluginContext.register_video_gen_provider() - hermes_cli/tools_config.py — Video Generation category in hermes tools, plugin-only providers list, model picker per plugin, config write to video_gen.{provider,model} - toolsets.py — new video_gen toolset - tests: 31 new tests covering ABC, registry, tool dispatch, both plugins - docs: developer-guide/video-gen-provider-plugin.md (parallel to the image-gen guide), sidebar + toolsets-reference + plugin guides updated Supersedes: NousResearch#25035 (FAL), NousResearch#17972 (FAL), NousResearch#14543 (xAI), NousResearch#13847 (HappyHorse), NousResearch#10458 (provider categories), NousResearch#10786 (xAI media+search bundle), NousResearch#2984 (FAL duplicate), NousResearch#19086 (Google Veo standalone — easy port to plugin interface). Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com> * feat(video_gen): dynamic schema reflects active backend's capabilities Address the 'capability variance' question — instead of one tool with a static schema that lies about what every backend supports, the video_generate tool now rebuilds its description at get_definitions() time based on the configured video_gen.provider and video_gen.model. The agent sees backend-specific guidance up-front: - 'fal-ai/veo3.1/image-to-video': 'image-to-video only — image_url is REQUIRED; text-only prompts will be rejected' - 'fal-ai/veo3.1' (t2v): no image_url restriction shown - xAI grok-imagine-video: 'operations: generate, edit, extend; up to 7 reference_image_urls' - Backends without edit/extend: 'not supported on this backend — surface that they need to switch backends via hermes tools' This is the same pattern PR NousResearch#22694 used for delegate_task self-capping — documented in the dynamic-tool-schemas skill. Cache invalidation is free: get_tool_definitions() already memoizes on config.yaml mtime, so a mid-session backend swap rebuilds the schema automatically. Tested: - Empirical FAL OpenAPI schema check confirms image-to-video models require image_url (FAL returns HTTP 422 otherwise) — client-side rejection in FALVideoGenProvider.generate() now prevents the wasted round-trip - Live E2E: fal-ai/veo3.1/image-to-video + prompt-only → clean missing_image_url error; fal-ai/veo3.1 + prompt-only → dispatches - 6 new tests cover the builder (no config / image-only / full-surface / text-only / unknown provider / registry wiring), all passing - 37/37 in the slice, 134/134 in the broader regression set * test(video_gen/xai): full surface integration tests + cleaner schema Verified end-to-end that the xAI plugin handles every documented mode from PR NousResearch#10600's surface: text-to-video, image-to-video, reference-images-to-video, video edit, video extend (with and without prompt). All five modes route to the correct xAI endpoint (/videos/generations, /videos/edits, /videos/extensions) with the right payload shape (image / reference_images / video keys), and all five client-side rejections fire before the network: edit-without-prompt, extend-without-video_url, image+refs conflict, >7 references, and duration/aspect_ratio clamping. 15 new integration tests grouped into four classes (endpoint routing, modalities, validation, clamping). httpx is stubbed via a small fake AsyncClient that records POSTs so the tests assert the actual payload the plugin would send to xAI — not just the success/error envelope. Also cleaned up a description redundancy: when a model's operations match the backend's overall set, we no longer print the duplicate 'operations supported by this model' line. xAI's description now reads: Active backend: xAI . model: grok-imagine-video - operations supported by this backend: edit, extend, generate - modalities supported by this backend: image, reference_images, text - aspect_ratio choices: 16:9, 1:1, 2:3, 3:2, 3:4, 4:3, 9:16 - resolution choices: 480p, 720p - duration range: 1-15s - reference_image_urls: up to 7 images Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com> * feat(video_gen): collapse surface to t2v + i2v, family-based auto-routing Two design changes per Teknium: 1) Drop edit/extend from the tool surface entirely. Only text-to-video and image-to-video remain. The agent sees a clean tool with two modalities; backend-specific quirks like xAI's edit/extend endpoints stay out of the unified schema. 2) FAL: pick a model FAMILY once, the plugin routes between the family's text-to-video and image-to-video endpoints based on whether image_url was passed. Users no longer pick 'fal-ai/veo3.1' AND 'fal-ai/veo3.1/image-to-video' as separate options — they pick 'veo3.1', and the plugin handles the rest. Catalog rewritten as families: veo3.1 fal-ai/veo3.1 / fal-ai/veo3.1/image-to-video pixverse-v6 fal-ai/pixverse/v6/text-to-video / fal-ai/pixverse/v6/image-to-video kling-o3-standard fal-ai/kling-video/o3/standard/text-to-video / fal-ai/kling-video/o3/standard/image-to-video xAI uses a single endpoint (/videos/generations) for both modes, routed by the presence of the 'image' field in the payload — no edit/extend exposure. Schema changes: - VIDEO_GENERATE_SCHEMA: drop operation, drop video_url. Final params: prompt (required), image_url, reference_image_urls, duration, aspect_ratio, resolution, negative_prompt, audio, seed, model. - VideoGenProvider ABC: drop normalize_operation, VALID_OPERATIONS, DEFAULT_OPERATION. capabilities() drops 'operations' key. - success_response: add 'modality' field ('text' | 'image') so the agent and logs can see which endpoint was actually hit. Dynamic schema builder simplified — no operations bullet, no 'switch backends if you need edit/extend' guidance. When the active backend supports both modalities (the common case), description reads: Active backend: FAL . model: pixverse-v6 - supports both text-to-video (omit image_url) and image-to-video (pass image_url) - routes automatically - aspect_ratio choices: 16:9, 9:16, 1:1 - resolution choices: 360p, 540p, 720p, 1080p - duration range: 1-15s - audio: pass audio=true to enable native audio (pricing tier) - negative_prompt: supported Tests: 51 in the video_gen slice, 216 across the broader image+video sweep, all passing. New FAL routing tests prove pixverse-v6 + no image hits text-to-video endpoint, pixverse-v6 + image_url hits image-to-video endpoint, same for veo3.1 and kling-o3-standard. Docs updated: developer-guide page rewrites the 'model families' pattern as a first-class section so external plugin authors know the convention. toolsets-reference and toolsets.py descriptions match the new surface. Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com> * feat(video_gen/fal): expand catalog to 6 families, cheap + premium tiers Catalog now covers everything Teknium specced from FAL: Cheap tier: ltx-2.3 fal-ai/ltx-2.3-22b/text-to-video / image-to-video pixverse-v6 fal-ai/pixverse/v6/text-to-video / image-to-video Premium tier: veo3.1 fal-ai/veo3.1 / fal-ai/veo3.1/image-to-video seedance-2.0 bytedance/seedance-2.0/text-to-video / image-to-video kling-v3-4k fal-ai/kling-video/v3/4k/text-to-video / image-to-video happy-horse fal-ai/happy-horse/text-to-video / image-to-video DEFAULT_MODEL moved from veo3.1 (premium) to pixverse-v6 (cheap, sane defaults, both modalities) — better first-run UX for users who haven't explicitly picked a model. New family-entry knob: image_param_key. Kling v3 4K's image-to-video endpoint expects start_image_url instead of image_url; declaring image_param_key='start_image_url' on the family lets _build_payload remap correctly. Other families default to plain image_url. Per-family capability flags reflect each model's docs: - LTX 2.3 + Happy Horse: minimal payloads (no duration/aspect/resolution enum exposed by FAL — let endpoint apply defaults) - Seedance: 6 aspect ratios incl 21:9, durations 4-15, audio supported, negative prompts NOT supported per docs - Kling v3 4K: 16:9/9:16/1:1, 3-15s, audio + negative - Veo 3.1: unchanged, 16:9/9:16, 4/6/8s Tests: +5 covering the new families (full catalog, Kling 4K start_image_url remap, Seedance routing, LTX payload minimality, Happy Horse minimality). 56/56 in the slice green. Note: I did NOT add the FAL-hosted xAI Grok-Imagine variant. Hermes already has a direct xAI plugin that talks to xAI's own API; routing the same model through FAL's wrapper would duplicate the surface without adding capabilities. Users on FAL who want Grok-Imagine should use the xAI plugin directly; flag if you want both routes available. * test(video_gen): tool-surface routing matrix — every model x modality End-to-end matrix test driven through _handle_video_generate() — the actual function the agent's video_generate tool call lands in. Writes config.yaml, invokes the registered handler with a raw args dict, then asserts the outbound HTTP/SDK call hit the right endpoint with the right payload shape. Parametrized over FAL_FAMILIES.keys() so the matrix auto-discovers new families as they're added (add a family to FAL_FAMILIES and you get both modalities tested for free). Coverage: - All 6 FAL families x {text-only, text+image} = 12 cases - xAI x {text-only, text+image} = 2 cases - tool-level model= arg overrides config = 2 cases For each case, verifies: - result['success'] is True - result['modality'] matches input shape ('text' if no image_url, 'image' otherwise) - outbound endpoint URL matches the family's text_endpoint or image_endpoint - text-only payloads carry no image-shaped keys - text+image payloads carry the family's image key (image_url for most, start_image_url for kling-v3-4k, wrapped 'image' object for xAI) All 16 cases passing. Confirms the tool surface routes every (provider, model, modality) combination correctly with zero leakage. * feat(video_gen): keep video_gen out of first-run setup, surface in status Two changes: 1. video_gen joins _DEFAULT_OFF_TOOLSETS, so it is NOT pre-selected in the first-run toolset checklist. Video gen is niche, paid, and slow — most users don't want it nagging them during initial setup. Anyone who wants it opts in via 'hermes tools' -> Video Generation, which already routes to the provider+model picker. 2. The 'hermes setup' status panel learns about video_gen — but only shows the row when a plugin reports available. Users without FAL_KEY/XAI_API_KEY see nothing about video gen; users with one of those keys see 'Video Generation (FAL) ✓' as confirmation it's wired. Verified live: - Fresh install (no creds): zero video_gen mentions in wizard. - With FAL_KEY: status row appears with active backend name. - 160/160 in the setup + tools_config + video_gen test slice. Rationale: image_gen is on by default because it's a featured creative tool used in casual chat (telegrams, etc). Video gen is heavier — long wait, paid per-second pricing. Default-off matches user intent better. --------- Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
…ex models (NousResearch#24182) * feat(codex-runtime): scaffold optional codex app-server runtime Foundational commit for an opt-in alternate runtime that hands OpenAI/Codex turns to a 'codex app-server' subprocess instead of Hermes' tool dispatch. Default behavior is unchanged. Lands in three pieces: 1. agent/transports/codex_app_server.py — JSON-RPC 2.0 over stdio speaker for codex's app-server protocol (codex-rs/app-server). Spawn, init handshake, request/response, notification queue, server-initiated request queue (for approval round-trips), interrupt-friendly blocking reads. Tested against real codex 0.130.0 binary end-to-end during development. 2. hermes_cli/runtime_provider.py: - Adds 'codex_app_server' to _VALID_API_MODES. - Adds _maybe_apply_codex_app_server_runtime() helper, called at the end of _resolve_runtime_from_pool_entry(). Inert unless 'model.openai_runtime: codex_app_server' is set in config.yaml AND provider in {openai, openai-codex}. Other providers cannot be rerouted (anthropic, openrouter, etc. preserved). 3. tests/agent/transports/test_codex_app_server_runtime.py — 24 tests covering api_mode registration, the rewriter helper (default-off, case-insensitive, opt-in, non-eligible providers preserved), version parser, missing-binary handling, error class. Does NOT require codex CLI installed. This commit is wire-only: the api_mode is recognized but AIAgent does not yet branch on it. Followup commits add the session adapter, event projector, approval bridge, transcript projection (so memory/skill review still works), plugin migration, and slash command. Existing tests remain green: - tests/cli/test_cli_provider_resolution.py (29 passed) - tests/agent/test_credential_pool_routing.py (included above) * feat(codex-runtime): add codex item projector for memory/skill review The translator that lets Hermes' self-improvement loop keep working under the Codex runtime: converts codex 'item/*' notifications into Hermes' standard {role, content, tool_calls, tool_call_id} message shape that agent/curator.py already knows how to read. Item taxonomy (matches codex-rs/app-server-protocol/src/protocol/v2/item.rs): - userMessage → {role: user, content} - agentMessage → {role: assistant, content: text} - reasoning → stashed in next assistant's 'reasoning' field - commandExecution → assistant tool_call(name='exec_command') + tool result - fileChange → assistant tool_call(name='apply_patch') + tool result - mcpToolCall → assistant tool_call(name='mcp.<server>.<tool>') + tool result - dynamicToolCall → assistant tool_call(name=<tool>) + tool result - plan/hookPrompt/etc → opaque assistant note, no fabricated tool_calls Invariants preserved: - Message role alternation never violated: each tool item produces at most one assistant + one tool message in that order, correlated by call_id. - Streaming deltas (item/<type>/outputDelta, item/agentMessage/delta) don't materialize messages — only item/completed does. Mirrors how Hermes already only writes the assistant message after streaming ends. - Tool call ids are deterministic (codex item id-based) so replays produce identical messages and prefix caches stay valid (AGENTS.md pitfall NousResearch#16). - JSON args use sorted_keys for the same reason. Real wire formats verified against codex 0.130.0 by capturing live notifications from thread/shellCommand and including one as a fixture (COMMAND_EXEC_COMPLETED). 23 new tests, all green: - Streaming deltas don't materialize (3 paths) - Turn/thread frame events are silent - commandExecution: 5 tests including non-zero exit annotation + deterministic id stability across replays - agentMessage + reasoning attachment + reasoning consumption - fileChange: summary without inlined content - mcpToolCall: namespaced naming + error surfacing - userMessage: text fragments only (drops images/etc) - opaque items: no fabricated tool_calls - Helpers: deterministic id stability + sorted JSON args - Role alternation invariant across all four tool-shaped item types This commit is a pure addition. AIAgent integration (the wire that uses the projector) is the next commit. * feat(codex-runtime): add session adapter + approval bridge The third self-contained module: CodexAppServerSession owns one Codex thread per Hermes session, drives turn/start, consumes streaming notifications via CodexEventProjector, handles server-initiated approval requests, and translates cancellation into turn/interrupt. The adapter has a single public per-turn method: result = session.run_turn(user_input='...', turn_timeout=600) # result.final_text → assistant text for the caller # result.projected_messages → list ready to splice into AIAgent.messages # result.tool_iterations → tick count for _iters_since_skill nudge # result.interrupted → True on Ctrl+C / deadline / interrupt # result.error → error string when the turn cannot complete # result.turn_id, thread_id → for sessions DB / resume Behavior: - ensure_started() spawns codex, does the initialize handshake, and issues thread/start with cwd + permissions profile. Idempotent. - run_turn() blocks until turn/completed, drains server-initiated requests (approvals) before reading notifications so codex never deadlocks waiting for us, projects every item/completed via the projector, and increments tool_iterations for the skill nudge gate. - request_interrupt() is thread-safe (threading.Event); the next loop iteration issues turn/interrupt and unwinds. - turn_timeout deadlock guard issues turn/interrupt and records an error if the turn never completes. - close() escalates terminate → kill via the underlying client. Approval bridge: Codex emits server-initiated requests for execCommandApproval and applyPatchApproval. The adapter translates Hermes' approval choice vocabulary onto codex's decision vocabulary: Hermes 'once' → codex 'approved' Hermes 'session' or 'always' → codex 'approvedForSession' Hermes 'deny' / anything else → codex 'denied' Routing precedence: 1. _ServerRequestRouting.auto_approve_* flags (cron / non-interactive) 2. approval_callback wired by the CLI (defers to tools.approval.prompt_dangerous_approval()) 3. Fail-closed denial when neither is wired Unknown server-request methods are answered with JSON-RPC error -32601 so codex doesn't hang waiting for us. Permission profile mapping mirrors AGENTS.md: Hermes 'auto' → codex 'workspace-write' Hermes 'approval-required' → codex 'read-only-with-approval' Hermes 'unrestricted/yolo' → codex 'full-access' 20 new tests, all green. Combined with prior commits this PR now has 67 tests across three modules: - test_codex_app_server_runtime.py: 24 (api_mode + transport surface) - test_codex_event_projector.py: 23 (item taxonomy projections) - test_codex_app_server_session.py: 20 (turn loop + approvals + interrupts) Full tests/agent/transports/ directory: 249/249 pass — no regressions to existing transport tests. Still no wire into AIAgent.run_conversation(); that integration commit is small and goes next. * feat(codex-runtime): wire codex_app_server runtime into AIAgent The integration commit. AIAgent.run_conversation() now early-returns to a new helper _run_codex_app_server_turn() when self.api_mode == 'codex_app_server', bypassing the chat_completions tool loop entirely. Three small surgical edits to run_agent.py (~105 LOC total): 1. Line ~1204 (constructor api_mode validation set): Add 'codex_app_server' so an explicit api_mode='codex_app_server' passed to AIAgent() isn't silently rewritten to 'chat_completions'. 2. Line ~12048 (run_conversation, just before the while loop): Early-return to _run_codex_app_server_turn() when self.api_mode is 'codex_app_server'. Placed AFTER all standard pre-loop setup — logging context, session DB, surrogate sanitization, _user_turn_count and _turns_since_memory increments, _ext_prefetch_cache, memory manager on_turn_start — so behavior outside the model-call loop is identical between paths. Default Hermes flow is unchanged when the flag is off. 3. End-of-class (line ~15497): New method _run_codex_app_server_turn(). Lazy-instantiates one CodexAppServerSession per AIAgent (reused across turns), runs the turn, splices projected_messages into messages, increments _iters_since_skill by tool_iterations (since the chat_completions loop normally does that per iteration), fires _spawn_background_review on the same cadence as the default path. Counter accounting: _turns_since_memory ← already incremented at run_conversation:11817 (gated on memory store configured) — codex helper does NOT touch it (would double-count). _user_turn_count ← already incremented at run_conversation:11793 — codex helper does NOT touch it. _iters_since_skill ← incremented in the chat_completions loop per tool iteration. Codex helper increments by turn.tool_iterations since the loop is bypassed. User message: ALREADY appended to messages by run_conversation pre-loop (line 11823) before the early-return reaches us. Helper does NOT append again. Regression test test_user_message_not_duplicated guards this. Approval callback wiring: Lazy-fetches tools.terminal_tool._get_approval_callback at session spawn time, passes to CodexAppServerSession. CLI threads with prompt_toolkit get interactive approvals; gateway/cron contexts get the codex-side fail-closed deny. Error path: Codex session exceptions become a 'partial' result with completed=False and a final_response that explicitly tells the user how to switch back: 'Codex app-server turn failed: ... Fall back to default runtime with /codex-runtime auto.' Same return-dict shape as the chat_completions path so all callers (gateway, CLI, batch_runner, ACP) work unchanged. 9 new integration tests in tests/run_agent/test_codex_app_server_integration.py: - api_mode='codex_app_server' is accepted on AIAgent construction - run_conversation returns the expected codex shape (final_response, codex_thread_id, codex_turn_id, completed, partial) - Projected messages are spliced into messages list - _iters_since_skill ticks per tool iteration - _user_turn_count delegated to standard flow (not double-counted) - User message appears exactly once (regression guard) - _spawn_background_review IS invoked (memory/skill review keeps working) - chat.completions.create is NEVER called (loop fully bypassed) - Session exception → partial result with /codex-runtime auto hint - Interrupted turn → partial result with error preserved Adjacent test runs confirm no regressions: - tests/run_agent/test_memory_nudge_counter_hydration.py: green - tests/run_agent/test_background_review.py: green - tests/run_agent/test_fallback_model.py: green - tests/agent/transports/: 249/249 green Still missing for full feature: /codex-runtime slash command, plugin migration helper, docs page, live e2e test gated on codex binary. Those are the remaining followup commits. * feat(codex-runtime): add /codex-runtime slash command (CLI + gateway) User-facing toggle for the optional codex app-server runtime. Follows the 'Adding a Slash Command (All Platforms)' pattern from AGENTS.md exactly: single CommandDef in the central registry → CLI handler → gateway handler → running-agent guard → all surfaces (autocomplete, /help, Telegram menu, Slack subcommands) update automatically. Surface: /codex-runtime — show current state + codex CLI status /codex-runtime auto — Hermes default runtime /codex-runtime codex_app_server — codex subprocess runtime /codex-runtime on / off — synonyms Files changed: hermes_cli/codex_runtime_switch.py (new): Pure-Python state machine shared by CLI and gateway. Parse args, read/write model.openai_runtime in the config dict, gate enabling behind a codex --version check (don't let users opt in to a runtime they have no binary for; print npm install hint instead). Returns a CodexRuntimeStatus dataclass that callers render however suits their surface. hermes_cli/commands.py: Single CommandDef entry, no aliases (codex-runtime is its own thing). cli.py: Dispatch in process_command() + _handle_codex_runtime() handler that delegates to the shared module and renders results via _cprint. gateway/run.py: Dispatch in _handle_message() + _handle_codex_runtime_command() that returns a string (gateway sends as message). On a successful change that requires a new session, _evict_cached_agent() forces the next inbound message to construct a fresh AIAgent with the new api_mode — avoids prompt-cache invalidation mid-session. gateway/run.py running-agent guard: /codex-runtime joins /model in the early-intercept block so a runtime flip mid-turn can't split a turn across two transports. Tests: tests/hermes_cli/test_codex_runtime_switch.py — 25 tests covering the state machine: arg parsing (10 cases incl. case-insensitive and synonyms), reading current runtime (5 cases incl. malformed configs), writing runtime (3 cases), apply() entry point covering read-only, no-op, codex-missing-blocked, codex-present-success, disable-no-binary-check, and persist-failure paths (8 cases). All green. Adjacent test suites confirm no regressions: - tests/hermes_cli/test_commands.py + test_codex_runtime_switch.py: 167/167 green - tests/agent/transports/: 283/283 green when combined with prior commits Still missing: plugin migration helper, docs page, live e2e test gated on codex binary. Followup commits. * feat(codex-runtime): auto-migrate Hermes MCP servers to ~/.codex/config.toml Translates the user's mcp_servers config from ~/.hermes/config.yaml into the TOML format codex's MCP client expects. Wired into the /codex-runtime codex_app_server enable path so users get their MCP tool surface in the spawned subprocess automatically. The migration runs on every enable. Failures are non-fatal — the runtime change still proceeds and the user gets a warning so they can fix the codex config manually. What translates (mapping verified against codex-rs/core/src/config/edit.rs): Hermes mcp_servers.<n>.command/args/env → codex stdio transport Hermes mcp_servers.<n>.url/headers → codex streamable_http transport Hermes mcp_servers.<n>.timeout → codex tool_timeout_sec Hermes mcp_servers.<n>.connect_timeout → codex startup_timeout_sec Hermes mcp_servers.<n>.cwd → codex stdio cwd Hermes mcp_servers.<n>.enabled: false → codex enabled = false What does NOT translate (warned + skipped per server): Hermes-specific keys (sampling, etc.) — codex's MCP client has no equivalent. Listed in the per-server skipped[] field of the report. What's NOT migrated (intentional): AGENTS.md — codex respects this file natively in its cwd. Hermes' own AGENTS.md (project-level) is already in the worktree, so codex picks it up without translation. No code needed. Idempotency design: All managed content lives between a 'managed by hermes-agent' marker and the next non-mcp_servers section header. _strip_existing_managed_block removes the prior managed region cleanly, preserving any user-added codex config (model, providers.openai, sandbox profiles, etc.) above or below. Files added: hermes_cli/codex_runtime_plugin_migration.py — pure-Python migration helper. Public API: migrate(hermes_config, codex_home=None, dry_run=False) returns MigrationReport with .migrated/.errors/ .skipped_keys_per_server. No external TOML dependency — minimal formatter handles strings/numbers/booleans/lists/inline-tables. tests/hermes_cli/test_codex_runtime_plugin_migration.py — 39 tests covering: - per-server translation (12): stdio/http/sse, cwd, timeouts, enabled flag, command+url precedence, sampling drop, unknown keys - TOML formatter (8): types, escaping, inline tables, error case - existing-block stripping (4): no marker, alone, with user content above, with user content below - end-to-end migrate() (8): empty, dry-run, round-trip, idempotent re-run, preserves user config, error reporting, invalid input, summary formatting Files changed: hermes_cli/codex_runtime_switch.py — apply() now calls migrate() in the codex_app_server enable branch. Migration failure logs a warning in the result message but does NOT fail the runtime change. Disable path (auto) explicitly skips migration. tests/hermes_cli/test_codex_runtime_switch.py — 3 new tests: test_enable_triggers_mcp_migration, test_disable_does_not_trigger_migration, test_migration_failure_does_not_block_enable. All 325 feature tests green: - tests/agent/transports/: 249 (incl. 67 new) - tests/run_agent/test_codex_app_server_integration.py: 9 - tests/hermes_cli/test_codex_runtime_switch.py: 28 (3 new) - tests/hermes_cli/test_codex_runtime_plugin_migration.py: 39 (new) * perf(codex-runtime): cache codex --version check within apply() Single /codex-runtime invocation could spawn 'codex --version' up to 3 times (state report, enable gate, success message). Each spawn is ~50ms, so the cumulative cost wasn't a crisis, but it was wasteful and turned a trivial slash command into something noticeably laggy on slower systems. Refactored to lazy-once via a closure over a nonlocal cache. First call spawns; subsequent calls in the same apply() reuse the result. Behavior unchanged — same return shape, same error handling, same install hint when codex is missing. Just one subprocess per call instead of three. Two regression-guard tests added: - test_binary_check_cached_within_apply: enable path → call_count == 1 - test_binary_check_cached_on_read_only_call: state-report path → call_count == 1 Total tests for /codex-runtime now 30 (was 28); all 143 codex-runtime tests still green. * fix(codex-runtime): correct protocol field names found via live e2e test Three real bugs caught only by running a turn end-to-end against codex 0.130.0 with a real ChatGPT subscription. Unit tests passed because they asserted on our own (incorrect) wire shapes; the wire format from codex-rs/app-server-protocol/src/protocol/v2/* is the source of truth and my initial reading of the README was incomplete. Bug 1: thread/start.permissions wire format Was sending {"profileId": "workspace-write"}. Real format per PermissionProfileSelectionParams enum (tagged union): {"type": "profile", "id": "workspace-write"} AND requires the experimentalApi capability declared during initialize. AND requires a matching [permissions] table in ~/.codex/config.toml or codex fails the request with 'default_permissions requires a [permissions] table'. Fix: stop overriding permissions on thread/start. Codex picks its default profile (read-only unless user configures otherwise), which matches what codex CLI users expect — they configure their default permission profile in ~/.codex/config.toml the standard way. Trying to be clever about profile selection broke every turn we tested. Live error before fix: 'Invalid request: missing field type' on every turn/start, even though our turn/start payload was correct — the field codex was complaining about was inside the permissions sub-object we shouldn't have been sending. Bug 2: server-request method names Was matching 'execCommandApproval' and 'applyPatchApproval'. Real names per common.rs ServerRequest enum: item/commandExecution/requestApproval item/fileChange/requestApproval item/permissions/requestApproval (new third method) Fix: match the documented names. Added handler for item/permissions/requestApproval that always declines — codex sometimes asks to escalate permissions mid-turn and silent acceptance would surprise users. Live symptom before fix: agent.log showed 'Unknown codex server request: item/commandExecution/requestApproval' and codex stalled because we replied with -32601 (unsupported method) instead of an approval decision. The agent reported back 'The write command was rejected' even though Hermes never showed the user an approval prompt. Bug 3: approval decision values Was sending decision strings 'approved'/'approvedForSession'/'denied'. Real values per CommandExecutionApprovalDecision enum (camelCase): accept, acceptForSession, decline, cancel (also AcceptWithExecpolicyAmendment and ApplyNetworkPolicyAmendment variants we don't currently use). Fix: rename _approval_choice_to_codex_decision return values; update auto_approve_* fallbacks; update fail-closed default from 'denied' to 'decline'. Test mapping table updated to match. Live test verified after fixes: $ hermes (with model.openai_runtime: codex_app_server) > Run the shell command: echo hermes-codex-livetest > .../proof.txt then read it back Approval prompt fired with 'Codex requests exec in <cwd>'. User chose 'Allow once'. Codex executed the command, wrote the file, read it back. Final response: 'Read back from proof.txt: hermes-codex-livetest'. File contents on disk match. agent.log confirms: codex app-server thread started: id=019e200e profile=workspace-write cwd=/tmp/hermes-codex-livetest/workspace All 20 session tests still green after wire-format updates. * fix(codex-runtime): correct apply_patch approval params + ship docs Live e2e revealed FileChangeRequestApprovalParams doesn't carry the changeset (just itemId, threadId, turnId, reason, grantRoot) — Codex's 'reason' field describes what the patch wants to do. Test config and display logic updated to use it. The first 'apply_patch (0 change(s))' display from the live test is now 'apply_patch: <reason>'. Adds website/docs/user-guide/features/codex-app-server-runtime.md covering enable/disable, prerequisites, approval UX, MCP migration behavior, permission profile delegation to ~/.codex/config.toml, known limitations, and the architecture diagram. Wired into the Automation category in sidebars.ts. Live e2e validation across the path matrix: ✓ thread/start handshake ✓ turn/start with text input ✓ commandExecution items + projection ✓ item/commandExecution/requestApproval → Hermes UI → response ✓ Approve once → command runs ✓ Deny → command rejected, codex falls back to read-only message ✓ Multi-turn (codex remembers prior turn's results) ✓ apply_patch via Codex's fileChange path ✓ item/fileChange/requestApproval → Hermes UI ✓ MCP server migration loads inside spawned codex (verified via 'use the filesystem MCP tool' prompt) ✓ /codex-runtime auto → codex_app_server toggle cycle ✓ Disable doesn't trigger migration ✓ Enable with codex CLI present succeeds + migrates ✓ Hermes-side interrupt path (turn/interrupt request issued cleanly even if codex finishes before the interrupt lands) Known live-validated limitations now documented in the docs page: - delegate_task subagents unavailable on this runtime - permission profile selection delegated to ~/.codex/config.toml - apply_patch approval prompt has no inline changeset (codex protocol doesn't expose it) 145/145 codex-runtime tests still green. * feat(codex-runtime): native plugin migration + UX polish (quirks 2/4/5/10/11) Major: migrate native Codex plugins (NousResearch#7 in OpenClaw's PR list) Discovers installed curated plugins via codex's plugin/list RPC and writes [plugins."<name>@<marketplace>"] entries to ~/.codex/config.toml so they're enabled in the spawned Codex sessions. This is the 'YouTube-video-worthy' bit Pash highlighted: when a user has google-calendar, github, etc. installed in their Codex CLI, those plugins activate automatically when they enable Hermes' codex runtime. Implementation: - hermes_cli/codex_runtime_plugin_migration.py: new _query_codex_plugins() helper spawns 'codex app-server' briefly and walks plugin/list. Returns (plugins, error) — failures are non-fatal so MCP migration still works. - render_codex_toml_section() now takes plugins + permissions args. - migrate() defaults: discover_plugins=True, default_permission_profile= 'workspace-write'. Explicit None on either disables that side. - _strip_existing_managed_block() now also strips [plugins.*] and [permissions]/[permissions.*] sections inside the managed block, so re-runs replace plugins cleanly without touching codex's own config. Quirk fixes: #2 Default permissions profile written on enable. Without this, Codex's read-only default kicks in and EVERY write triggers an approval prompt. Now writes [permissions] default = 'workspace-write' so the runtime feels normal out of the box. Set default_permission_profile=None to opt out. NousResearch#4 apply_patch approval prompt now shows what's changing. Codex's FileChangeRequestApprovalParams doesn't carry the changeset. Session adapter now caches the fileChange item from item/started notifications and looks it up by itemId when codex requests approval. Prompt shows '1 add, 1 update: /tmp/new.py, /tmp/old.py' instead of 'apply_patch (0 change(s))'. Side benefit: also drains pending notifications BEFORE handling a server request, so the projector and per-turn caches are up to date when the approval decision fires. Bounded to 8 notifications per loop iter to avoid starving codex's response. NousResearch#5/NousResearch#10 Exec approval prompt never shows empty cwd. When codex omits cwd in CommandExecutionRequestApprovalParams, fall back to the session's cwd. If somehow neither is available, show '<unknown>' explicitly instead of an empty string. Also surfaces 'reason' from the approval params when codex provides it — gives users more context on why codex wants to run something. NousResearch#11 Banner indicates the codex_app_server runtime when active. New 'Runtime: codex app-server (terminal/file ops/MCP run inside codex)' line appears in the welcome banner only when the runtime is on. Default banner is unchanged. Tests: - 7 new tests in test_codex_runtime_plugin_migration.py covering plugin discovery (mocked), failure handling, dry-run skip, opt-out flag, idempotent re-runs, and permissions writing. - 3 new tests in test_codex_app_server_session.py covering the enriched approval prompts: cwd fallback, change summary on apply_patch, fallback when no item/started cache exists. - All 26 session tests + 46 migration tests green; 153 total in PR. * feat(codex-runtime): hermes-tools MCP callback + native plugin migration The big architectural addition: when codex_app_server runtime is on, Hermes registers its own tool surface as an MCP server in ~/.codex/config.toml so the codex subprocess can call back into Hermes for tools codex doesn't ship with — web_search, browser_*, vision, image_generate, skills, TTS. Also: 'migrate native codex plugins' (Pash's YouTube-video-worthy bit) — when the user has plugins like Linear, GitHub, Gmail, Calendar, Canva installed via 'codex plugin', Hermes discovers them via plugin/list and writes [plugins.<name>@openai-curated] entries so they activate automatically. New module: agent/transports/hermes_tools_mcp_server.py FastMCP stdio server exposing 17 Hermes tools. Each call dispatches through model_tools.handle_function_call() — same code path as the Hermes default runtime. Run with: python -m agent.transports.hermes_tools_mcp_server [--verbose] Exposed: web_search, web_extract, browser_navigate / _click / _type / _press / _snapshot / _scroll / _back / _get_images / _console / _vision, vision_analyze, image_generate, skill_view, skills_list, text_to_speech. NOT exposed (deliberately): - terminal/shell/read_file/write_file/patch — codex has built-ins - delegate_task/memory/session_search/todo — _AGENT_LOOP_TOOLS in model_tools.py:493, require running AIAgent context. Documented as a limitation and surfaced in the slash command output. Migration changes (hermes_cli/codex_runtime_plugin_migration.py): - _query_codex_plugins() spawns 'codex app-server' briefly to walk plugin/list and pull installed openai-curated plugins. Failures are non-fatal — MCP migration still completes. - render_codex_toml_section() now takes plugins + permissions args AND wraps the managed block with a MIGRATION_END_MARKER comment so the stripper can reliably find both ends, even when the block contains top-level keys (default_permissions = ...). - migrate() defaults: discover_plugins=True, expose_hermes_tools=True, default_permission_profile=':workspace' (built-in codex profile name — must be prefixed with ':'). All three opt-out via explicit args. - _build_hermes_tools_mcp_entry() builds the codex stdio entry with HERMES_HOME and PYTHONPATH passthrough so a worktree-launched Hermes points the MCP subprocess at the same module layout. Live-caught wire bugs fixed during this turn: 1. Permission profile config key is top-level , NOT a [permissions] table. The [permissions] table is for *user-defined* profiles with structured fields. Built-in profile names start with ':' (':workspace', ':read-only', ':danger-no-sandbox'). Was emitting which codex rejected with 'invalid type: string "X", expected struct PermissionProfileToml'. 2. Built-in profile is , NOT . Codex rejected with 'unknown built-in profile'. 3. Codex's MCP layer sends for tool-call confirmation. We weren't handling it, so codex stalled and returned 'MCP tool call was rejected'. Now: auto-accept for our own hermes-tools server (user already opted in by enabling the runtime), decline for third-party servers. Quirk fixes shipped (from the limitations list): #2 default permissions: workspace profile written on enable. No more approval prompt on every write. NousResearch#4 apply_patch approval shows what's changing: cache fileChange items from item/started, look up by itemId when codex sends item/fileChange/requestApproval. Prompt: '1 add, 1 update: /tmp/new.py, /tmp/old.py' instead of '0 change(s)'. NousResearch#5/NousResearch#10 exec approval cwd never empty: fall back to session cwd, then '<unknown>'. Also surfaces 'reason' from codex when present. NousResearch#11 banner shows 'Runtime: codex app-server' line when active so users understand why tool counts may not match what's reachable. Tests: - 5 new tests in test_codex_runtime_plugin_migration.py covering plugin discovery, expose_hermes_tools entry generation, idempotent re-runs, opt-out flag, permissions profile. - 3 new tests in test_codex_app_server_session.py covering enriched approval prompts (cwd fallback, fileChange summary). - 2 new tests for mcpServer/elicitation/request handling (accept hermes-tools, decline others). - New test file test_hermes_tools_mcp_server.py covering module surface, EXPOSED_TOOLS safety invariants (no shell/file_ops, no agent-loop tools), and main() error paths. - 166 codex-runtime tests total, all green. Live e2e validated against codex 0.130.0 + ChatGPT subscription: ✓ /codex-runtime codex_app_server enables, migrates filesystem MCP, registers hermes-tools, writes default_permissions = ':workspace' ✓ Banner shows 'Runtime: codex app-server' line in subsequent sessions ✓ Shell command runs without approval prompt (workspace profile works) ✓ Multi-turn — codex remembers prior turn's results ✓ apply_patch path via fileChange request approval ✓ web_search via hermes-tools MCP callback returns real Firecrawl results: 'OpenAI Codex CLI – Getting Started' end-to-end in 13s ✓ Disable cycle clean Docs updated: website/docs/user-guide/features/codex-app-server-runtime.md Full re-write covering native plugin migration, the hermes-tools callback architecture, the prerequisites change ('codex login is separate from hermes auth login codex'), the trade-off table now reflecting which Hermes tools work via callback, and the limitations list updated with what's actually unavailable on this runtime. * feat(codex-runtime): pin user-config preservation invariant for quirk NousResearch#6 Quirk NousResearch#6 from the limitations list — user MCP servers / overrides / codex-only sections in ~/.codex/config.toml that live OUTSIDE the hermes-managed block must survive re-migration verbatim. This already worked thanks to the MIGRATION_MARKER + MIGRATION_END_MARKER pair I added when fixing the default_permissions wire format (so the strip can find both ends of the managed region even with top-level keys like default_permissions). But it was an emergent property without a test pinning it. Now explicitly tested: - User MCP server above the managed block survives migration - User MCP server below the managed block survives migration - Both above + below survive a second re-migration - User content (model, providers, sandbox, otel, etc.) outside our region is left untouched Docs added a section "Editing ~/.codex/config.toml safely" explaining the marker contract — so users know they can add their own MCP servers, override permissions, configure codex-only options, etc. without fear of Hermes overwriting their work. 167 codex-runtime tests, all green. * docs(codex-runtime): clarify the actual tool surface — shell covers terminal/read/write/find Previous docs and PR description undersold what codex's built-in toolset actually provides. apply_patch alone made it sound like the runtime could only edit files in patch format — implying you'd lose terminal use, read_file, write_file, search/find. That was wrong. Codex's 'shell' tool runs arbitrary shell commands inside the sandbox, which covers everything you'd do in bash: cat/head/tail (read), echo> or heredocs (write), find/rg/grep (search), ls/cd (navigate), build/ test/git/etc. apply_patch is for structured multi-file edits on top of that. update_plan is its in-runtime todo. view_image loads images. And codex has its own web_search built in (in addition to the Firecrawl-backed one Hermes exposes via MCP callback). Docs now have a 'What tools the model actually has' section right after Why, breaking the surface into three clearly-labeled buckets: 1. Codex's built-in toolset (always on) — shell, apply_patch, update_plan, view_image, web_search; covers everything terminal- adjacent. 2. Native Codex plugins (auto-migrated from your codex plugin install) — Linear, GitHub, Gmail, Calendar, Outlook, Canva, etc. 3. Hermes tool callback (MCP server in ~/.codex/config.toml) — web_search/web_extract via Firecrawl, browser_*, vision_analyze, image_generate, skill_view/skills_list, text_to_speech. Plus a 'What's NOT available' callout listing the four agent-loop tools (delegate_task, memory, session_search, todo) that need running AIAgent context and can't reach the codex runtime. Trade-offs table broken out: shell, apply_patch, update_plan, view_image, sandbox each get their own row with a one-line description so users can see at a glance what's available natively. Architecture diagram updated to list the codex built-ins by name instead of 'apply_patch + shell + sandbox'. No code changes — purely docs clarification. 167 codex-runtime tests still green. * fix(codex-runtime): _spawn_background_review signature + review fork api_mode downgrade Two real bugs in the self-improvement loop integration that the previous test mocked away. Bug 1: wrong call signature The codex helper was calling self._spawn_background_review() with no args after every turn. That function actually requires: messages_snapshot=list (positional or keyword) review_memory=bool (at least one trigger must be True) review_skills=bool So the call would have raised TypeError at runtime — except the only test that exercised this path mocked _spawn_background_review entirely and just asserted spawn.called, so the wrong-arg shape never surfaced. Bug 2: review fork inherits codex_app_server api_mode The review fork is constructed with: api_mode = _parent_runtime.get('api_mode') So when the parent is codex_app_server, the review fork ALSO runs as codex_app_server. But the review fork's whole job is to call agent-loop tools (memory, skill_manage) which require Hermes' own dispatch — they short-circuit with 'must be handled by the agent loop' on the codex runtime. So the review fork would have run, decided to save something, called memory or skill_manage, and silently no-op'd. Fixed in run_agent.py:_spawn_background_review() — when the parent api_mode is 'codex_app_server', the review fork is downgraded to 'codex_responses' (same OAuth credentials, same openai-codex provider, but talks to OpenAI's Responses API directly so Hermes owns the loop). Also rewrote the codex helper's review wiring to match the chat_completions path: - Computes _should_review_memory in the pre-loop block (was already being computed; now passed through to the helper as an arg). - Computes _should_review_skills AFTER the codex turn returns + counters tick (line ~15432 pattern in chat_completions). - Calls _spawn_background_review(messages_snapshot=, review_memory=, review_skills=) only when at least one trigger fires. - Adds the external memory provider sync (_sync_external_memory_for_turn) that the chat_completions path runs after every turn. Tests: Replaced the broken test_background_review_invoked (which only asserted spawn.called) with three sharper tests: - test_background_review_NOT_invoked_below_threshold: single turn at default thresholds → no review fires (would have caught the original 'every turn calls spawn with no args' bug) - test_background_review_skill_trigger_fires_above_threshold: 10 tool_iterations at threshold=10 → review fires with messages_snapshot=list, review_skills=True, counter resets - test_background_review_signature_never_breaks: regression guard asserting positional args are always empty and kwargs include messages_snapshot New TestReviewForkApiModeDowngrade class: - test_codex_app_server_parent_downgrades_review_fork: drives the real _spawn_background_review function (no mock at that level), asserts the review_agent gets api_mode='codex_responses' when the parent was codex_app_server. Live-validated against real run_conversation: - Counter ticked from 0 to 5 after a 5-tool-iteration turn - _spawn_background_review fired exactly once with kwargs-only signature - review_skills=True, review_memory=False - messages_snapshot was 12 entries (5 assistant tool_calls + 5 tool results + 1 final assistant + initial system/user) - Counter reset to 0 after fire 170 codex-runtime tests, all green. Docs: added a Self-improvement loop section to the codex runtime page explaining both how the trigger logic stays equivalent and that the review fork is auto-downgraded to codex_responses for the agent-loop tools. Also clarified that apply_patch and update_plan ARE codex's built-in tools (the previous version made it sound like they were separate from 'codex's stuff' — they're not, all five tools listed in 'What tools the model actually has' section 1 are codex built-ins). * feat(codex-runtime): expose kanban tools through Hermes MCP callback Kanban workers spawn as separate hermes chat -q subprocesses that read the user's config.yaml. If model.openai_runtime: codex_app_server is set globally (which is the whole point of opt-in), every dispatched worker ALSO comes up on the codex runtime. That mostly works — codex's built-in shell + apply_patch + update_plan do the actual task work fine — but it had one critical break: the worker handoff tools (kanban_complete, kanban_block, kanban_comment, kanban_heartbeat) are Hermes-registered tools, not codex built-ins. On the codex runtime, codex builds its own tool list and these never reach the model, so the worker would do the work but not be able to report back, hanging until the dispatcher's timeout escalates it as zombie. Fix: add all 9 kanban tools to the EXPOSED_TOOLS list in the Hermes MCP callback. They dispatch statelessly through handle_function_call() just like web_search and the others — they read HERMES_KANBAN_TASK from env (set by the dispatcher), gate correctly (worker tools require the env var, orchestrator tools require it unset), and write to ~/.hermes/kanban.db. Why kanban tools work via stateless dispatch when delegate_task/memory/ session_search/todo don't: those four are listed in _AGENT_LOOP_TOOLS (model_tools.py:493) and short-circuit in handle_function_call() with 'must be handled by the agent loop' — they need to mutate AIAgent's mid-loop state. Kanban tools have no such requirement; they're pure side-effect functions against the kanban.db plus state_meta. Tools exposed: Worker handoff (require HERMES_KANBAN_TASK): kanban_complete, kanban_block, kanban_comment, kanban_heartbeat Read-only board queries: kanban_show, kanban_list Orchestrator (require HERMES_KANBAN_TASK unset): kanban_create, kanban_unblock, kanban_link Tests: - test_kanban_worker_tools_exposed: complete/block/comment/heartbeat in EXPOSED_TOOLS (regression guard for the would-hang-worker bug) - test_kanban_orchestrator_tools_exposed: create/show/list/unblock/link Docs: - New 'Workflow features' section in the docs page covering /goal, kanban, and cron behavior on this runtime - /goal: works fully via run_conversation feedback; only caveat is approval-prompt noise on long writes-heavy goals (mitigated by the default :workspace permission profile) - Kanban: enumerated which tools are reachable via the callback and why the env var propagates correctly through the codex subprocess to the MCP server subprocess - Cron: documented as 'not specifically tested' — same rules as the CLI apply since cron runs through AIAgent.run_conversation - Trade-offs table gained rows for /goal, kanban worker, kanban orchestrator 172/172 codex-runtime tests green (+2 from kanban tests). * docs(codex-runtime): wire /codex-runtime into slash-commands ref + flag aux token cost Three docs gaps caught during a final audit: 1. /codex-runtime was only in the feature docs page, not in the slash-commands reference. Added rows to both the CLI section and the Messaging section so users discover it where they'd look for slash command syntax. 2. CODEX_HOME and HERMES_KANBAN_TASK weren't in environment-variables.md. CODEX_HOME lets users redirect Codex CLI's config dir (the migration honors it). HERMES_KANBAN_TASK is set by the kanban dispatcher and propagates to the codex subprocess + the hermes-tools MCP subprocess so kanban worker tools gate correctly — documented as 'don't set manually' since it's an internal handoff. 3. Aux client behavior on this runtime. When openai_runtime= codex_app_server is on with the openai-codex provider, every aux task (title generation, context compression, vision auto-detect, session search summarization, the background self-improvement review fork) flows through the user's ChatGPT subscription by default. This is true for the existing codex_responses path too, but it's more visible / important here because users explicitly opted in for subscription billing. Added a 'Auxiliary tasks and ChatGPT subscription token cost' section to the docs page with a YAML example showing how to override specific aux tasks to a cheaper model (typically google/gemini-3-flash-preview via OpenRouter). Also documents how the self-improvement review fork gets auto-downgraded from codex_app_server to codex_responses by the fix earlier in this PR. No code changes — pure docs. 172 codex-runtime tests still green. * docs+test(codex-runtime): pin HOME passthrough, document multi-profile + CODEX_HOME OpenClaw hit a real footgun in openclaw/openclaw#81562: when spawning codex app-server they were synthesizing a per-agent HOME alongside CODEX_HOME. That made every subprocess codex's shell tool launches (gh, git, aws, npm, gcloud, ...) see a fake $HOME and miss the user's real config files. They had to back it out in PR NousResearch#81562 — keep CODEX_HOME isolation, leave HOME alone. Audit confirms Hermes' codex spawn doesn't have this problem. We do os.environ.copy() and only overlay CODEX_HOME (when provided) and RUST_LOG. HOME passes through unchanged. But it was an emergent property without a test pinning it, so adding a regression guard: test_spawn_env_preserves_HOME — confirms parent HOME survives intact in the subprocess env test_spawn_env_sets_CODEX_HOME_when_provided — confirms codex_home arg still isolates codex state correctly Docs additions: 'HOME environment variable passthrough' section — calls out the contract explicitly: CODEX_HOME isolates codex's own state, HOME stays user-real so gh/git/aws/npm/etc. find their normal config. Cites openclaw#81562 as the cautionary tale. 'Multi-profile / multi-tenant setups' section — addresses the related concern: profiles share ~/.codex/ by default. For users who want per-profile codex isolation (separate auth, separate plugins), documents the manual CODEX_HOME=<profile-scoped-dir> approach. Explains why we DON'T auto-scope CODEX_HOME per profile: doing so would silently invalidate existing codex login state for anyone upgrading to this PR with tokens already at ~/.codex/auth.json. Opt-in is safer than surprising users. 174 codex-runtime tests (+2 from HOME guards), all green. * fix(codex-runtime): TOML control-char escapes + atomic config.toml write Two footguns caught in a final audit pass before merge. Bug 1: TOML control characters not escaped The _format_toml_value() helper escaped backslashes and double quotes but passed literal control characters (\n, \t, \r, \f, \b) through unchanged. TOML basic strings don't allow literal control characters — a path or env var containing a newline would produce invalid TOML that codex refuses to load. Realistic exposure: pathological cases like a HERMES_HOME with a trailing newline (env var concatenation accident), or a PYTHONPATH with a tab from a multi-line shell heredoc. Fix: escape all five TOML basic-string control sequences (\b \t \n \f \r) in addition to \\ and \" that we already did. Order matters — backslash must come first or the other escapes get re-escaped. Bug 2: config.toml write wasn't atomic If the python process crashed between target.mkdir() and the write_text() finishing, a half-written config.toml could be left behind. On NFS / Windows / some FUSE mounts this is a real concern; on ext4/APFS small writes are usually atomic in practice but not guaranteed. Fix: write to a tempfile.mkstemp() temp file in the same directory, then Path.replace() (atomic same-dir rename on POSIX, ReplaceFile on Windows). On rename failure, clean up the temp file so repeated failed migrations don't pile up .config.toml.* files. Tests: - test_string_with_newline_escaped — \n in value → \n in output - test_string_with_tab_escaped — \t in value → \t in output - test_string_with_other_controls_escaped — \r, \f, \b - test_windows_path_escaped_correctly — backslash doubling - test_atomic_write_no_temp_leak_on_success — no .config.toml.* left over after a successful write - test_atomic_write_cleanup_on_rename_failure — temp file removed when Path.replace raises (simulated disk full) 180 codex-runtime tests, all green (+6 from this commit). Footguns audited but NOT fixed (with rationale): - Concurrent migrations race. Two Hermes processes hitting /codex-runtime codex_app_server within seconds of each other could cause one writer to lose entries. Low probability (you'd have to enable from two surfaces simultaneously) and low impact (just re-run migration). Adding fcntl/msvcrt locking is more code than it's worth here. The atomic rename above means each individual write is consistent — only the merge step is racy. - Codex protocol version drift. We pin MIN_CODEX_VERSION=0.125 and check at runtime but don't reject too-new versions. Right call — the protocol has been stable through 0.125 → 0.130. If OpenAI breaks it later we'd see the error in test_codex_app_server_runtime on CI before users hit it.
Adds a comprehensive EVM blockchain skill with 14 commands: - stats, wallet, tx, token, activity, gas, price (core queries) - compare: gas + prices across all 8 chains simultaneously - whale: scan recent blocks for large transfers (configurable min USD) - multichain: scan same wallet across all 8 chains in parallel - allowance: check dangerous ERC-20 approvals (Permit2, Uniswap, 1inch...) - decode: decode tx input data via 4byte.directory - ens: resolve ENS names <-> addresses (bidirectional) - contract: inspect contracts (proxy detection, ERC-20/721, bytecode size) Chains: Ethereum, BNB Chain, Base, Arbitrum One, Polygon, Optimism, Avalanche, zkSync Era Zero external dependencies. Python stdlib only (urllib, json, argparse, threading). Co-authored-by: Mibayy <mibay@clawhub.io>
…ousResearch#2010 Salvages the closed PR NousResearch#2010 (Mibayy's EVM multi-chain skill) and folds the existing optional-skills/blockchain/base/ skill into it, so we ship one unified EVM skill instead of two overlapping ones. Pulled in from base/: - 8 missing Base-specific tokens (AERO, DEGEN, TOSHI, BRETT, WELL, cbETH, cbBTC, wstETH, rETH) added to KNOWN_TOKENS['base'] — base/ had 11, evm/ only had 3 (USDC/DAI/WETH). - L1 data-fee pitfall note for rollups (Base, Arbitrum, Optimism, zkSync). - Batch-size chunking in rpc_batch (Base RPC caps batches at 10 calls per JSON-RPC request; adding more known tokens tripped that limit and broke 'wallet --chain base' with a 'list index out of range' error). Ported the chunking pattern from base/_rpc_batch_chunk. Latent bugs found and fixed while smoke-testing the merge: - cmd_multichain and cmd_allowance both iterated KNOWN_TOKENS[chain] with 'for contract, (symbol, _name) in known.items()' — but the dict shape is {symbol: contract_str}, not {addr: (sym, name)}. This raised 'too many values to unpack (expected 2)' on every non-zero balance. Now iterates as 'for symbol, contract in known.items()'. - Input validation: added is_valid_address / is_valid_txhash / require_address / require_txhash helpers and wired them into cmd_wallet, cmd_tx, cmd_token, cmd_activity, cmd_allowance, cmd_decode, cmd_contract, cmd_multichain. Fails fast with exit 2 on malformed input instead of burning an RPC round-trip on garbage. Documentation: - SKILL.md now flags that this skill supersedes optional-skills/blockchain/base. - Pitfalls expanded for ENS (single-endpoint dependency on ensideas.com), tx decoding (single-endpoint dependency on 4byte.directory), and rollup L1 fees. - Regenerated website/docs/user-guide/skills/optional/blockchain/ blockchain-evm.md and removed the old blockchain-base.md page; catalog updated. Removed: - optional-skills/blockchain/base/SKILL.md - optional-skills/blockchain/base/scripts/base_client.py - website/docs/user-guide/skills/optional/blockchain/blockchain-base.md Smoke-tested live against Base mainnet: stats, price, token, wallet (vitalik.eth — 3.12 ETH + 13.88 USDC + 4.23 DAI + 0.06 WETH on Base) and allowance (ethereum, 7 unlimited approvals to Uniswap/Permit2). Original PR NousResearch#2010 author: Mibayy. Original base/ skill author: youssefea.
- description ≤60 chars (was 346) - platforms: [linux, macos, windows] — script is pure stdlib (urllib, json, argparse), no POSIX-only primitives - author: credit @Mibayy + @youssefea + @ethernet8023 + Hermes Agent (was just Mibayy) - regenerated auto-gen docs page
…rds (NousResearch#25302) AGENTS.md and CONTRIBUTING.md both now state: 1. No new memory providers in the repo. The set under plugins/memory/ (honcho, mem0, supermemory, byterover, hindsight, holographic, openviking, retaindb) is closed. New backends ship as standalone plugin repos that users install into ~/.hermes/plugins/ via the same MemoryProvider ABC, discovery path, and hermes memory setup integration. PRs adding a new plugins/memory/<name>/ directory get closed with a pointer to publish as their own repo. 2. Skill authoring standards (hardline) — applies to all new or modernized skills (bundled, optional, contributed): - description <= 60 chars, one sentence, ends with period, no marketing words, no name repetition (verification snippet included) - tools referenced in SKILL.md prose must be native Hermes tools or MCP servers the skill expects — no grep/cat/sed/find etc. when search_files/read_file/patch already cover them - platforms: gating audited against actual POSIX-only primitives - author credits the human contributor first, not 'Hermes Agent' - SKILL.md uses modern section order with line targets - scripts/references/templates layout for non-trivial logic - tests at tests/skills/test_<skill>_skill.py, stdlib + mock only - .env.example edits isolated to a delimited block CONTRIBUTING.md includes a good/bad description example and a 'don't say / say' table mapping shell utilities to native tools. AGENTS.md points the agent at references/new-skill-pr-salvage.md for the full salvage checklist.
…ousResearch#25307) `_reconfigure_provider()` handled `image_gen_plugin_name` in both branches (no-env-vars early return and post-env-vars) but never mirrored the same handling for `video_gen_plugin_name`. The first-time `_configure_provider()` path correctly routes to `_select_plugin_video_gen_provider()`; reconfigure forgot to. Repro: 1. Enable video_gen in `hermes tools` → Configure for All Platforms. 2. Go back into `hermes tools` → Reconfigure tool → Video Generation. 3. Pick xAI (with XAI_API_KEY already set). 4. Hit Enter at the "keep current key?" prompt. Expected: `video_gen.provider: xai` written to config.yaml. Actual: function returns silently; no `video_gen:` block ever written; `video_generate` tool fails with "No video generation backend is configured." Fix: add the missing `video_gen_plugin_name` branch in both code paths of `_reconfigure_provider()`, mirroring the existing `image_gen_plugin_name` handling and the first-time configure logic. Tests: `tests/hermes_cli/test_video_gen_picker.py` covers both branches (env-vars-set keep-current and no-env-vars paths).
…`hermes tools` menus (NousResearch#25341) `hermes tools` -> "All Platforms" took ~14s to render the checklist because building the toolset labels called `get_nous_auth_status()` ~31x transitively (`_toolset_has_keys` -> `_visible_providers` -> `get_nous_subscription_features` -> `managed_nous_tools_enabled`). Each call did a synchronous OAuth refresh POST to portal.nousresearch.com (~350ms even on the failure path), so one menu paint burned >13s of HTTP and 31 single-use Nous refresh tokens. Secondary hot spot: every `get_env_value()` re-read and re-sanitised the entire .env file. 116 reads with O(lines x known-keys) scanning added ~300ms of CPU per render. Fix is two process-level caches, both mtime-keyed so login/logout/edit invalidate naturally: * `hermes_cli/auth.py`: memoise `get_nous_auth_status()` for 15s keyed on auth.json mtime. Splits `_compute_nous_auth_status()` as the uncached impl. Adds `invalidate_nous_auth_status_cache()`. * `hermes_cli/config.py`: memoise `load_env()` keyed on .env (path, mtime, size). Adds `invalidate_env_cache()`, wired into `save_env_value`, `remove_env_value`, and the sanitize-on-load writer so writers don't return stale dicts on same-second writes. Before/after on Teknium's box (real HERMES_HOME, no Nous login): * "All Platforms" cold path: ~13,874ms -> ~691ms label-build * Warm re-open within the same process: ~122ms -> ~17ms Side benefit: stops burning a Nous refresh token on every menu paint, which was risking the portal's reuse-detection revocation logic.
…hreads (NousResearch#25355) Slack platform-blocks native slash commands inside thread replies ("/queue is not supported in threads. Sorry!") and there is no app-side setting to re-enable them. As a workaround, rewrite a leading '!' to '/' for any known gateway command before downstream processing — so '!queue', '!stop', '!model gpt-5.4' etc. work inside Slack threads (and anywhere else). Only the first token is checked against is_gateway_known_command(), so casual messages like '!nice work' pass through to the agent unchanged. Downstream pipeline (MessageType.COMMAND tagging, gateway dispatcher, thread reply routing) is unchanged. Adds 6 tests covering rewrite, args preservation, thread routing, casual-message passthrough, '@bot' suffix, and plain '/' still-works.
…ndpoint The MiniMax TTS defaults were outdated: - DEFAULT_MINIMAX_MODEL was 'speech-01' but MiniMax now uses 'speech-02' - DEFAULT_MINIMAX_BASE_URL was 'https://api.minimax.chat/v1/text_to_speech' which no longer works; the correct endpoint is 'https://api.minimaxi.com/v1/t2a_v2' Users who configured tts.provider: minimax were getting model-not-supported errors because the hardcoded defaults did not match available API permissions.
… support Follow-up on @pty819's t2a_v2 endpoint fix: - Default model: speech-02 -> speech-02-hd (bare 'speech-02' is not in the supported enum; t2a_v2 rejects it with 400). Official enum: speech-01-hd, speech-01-turbo, speech-02-hd, speech-02-turbo, speech-2.6-hd/turbo, speech-2.8-hd/turbo. - Default voice: female-shaonv -> English_expressive_narrator. The legacy speech-01-series short ID doesn't resolve cleanly on the speech-02+ models that are now the default. - Default base URL: api.minimaxi.com -> api.minimax.io (matches the canonical host in the published docs; api-uw.minimax.io is the reduced-latency alt). - Add GroupId support via tts.minimax.group_id config or MINIMAX_GROUP_ID env var. Some MiniMax accounts scope TTS requests by group; without it, requests 401. Only appended when not already in the user's base_url. Tests rewritten to cover both the default t2a_v2 path (hex-encoded audio in JSON, nested voice_setting/audio_setting) and the legacy text_to_speech path (raw audio bytes, flat payload). Adds coverage for GroupId config/env wiring and error surfacing. Also adds AUTHOR_MAP entry for pty819's GitHub-noreply email.
…add ensure_and_bind helper (NousResearch#25038) Fixes NousResearch#25028. The lazy-install hooks added in NousResearch#25014 installed packages correctly but failed to rebind module-level globals after install: - Slack: missing aiohttp rebind → NameError on file uploads - Feishu: none of the ~25 lark_oapi symbols rebound → TypeError on adapter instantiation - Matrix: mautrix.types enums stayed as stubs → mismatched values at runtime Introduces tools.lazy_deps.ensure_and_bind() — a DRY helper that combines ensure() + importer-callable + globals().update(). This eliminates the error-prone pattern of manually listing every global that needs updating after lazy-install. Each platform adapter now defines a single _import() function returning all bindings. Also fixes: pyproject.toml [slack] extra was missing aiohttp (needed by slack-bolt's async path).
Adds set_thread_tool_whitelist / clear_thread_tool_whitelist to hermes_cli/plugins.py. When set on the current thread, restricts which tools can pass through get_pre_tool_call_block_message; non-whitelisted tools are blocked with a configurable deny message. Mirrors the per-thread approval-callback pattern already used by set_approval_callback (tools/terminal_tool.py:190). Used by _spawn_background_review to deny non-memory/non-skill tools at runtime while inheriting the parent agent's full tools schema for prefix-cache parity (see follow-up commit). Tests cover allow / deny / clear / cross-thread isolation. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… not skipped When the final streamed text is identical to the last plain-text edit, stream_consumer._send_or_edit short-circuits and never calls adapter.edit_message(finalize=True). For Telegram, this skips the plain-text → MarkdownV2 conversion, leaving raw Markdown syntax visible to the user. Set REQUIRES_EDIT_FINALIZE = True on TelegramAdapter so the finalize edit is always delivered, matching the existing DingTalk pattern. Fixes NousResearch#25710
`hermes config set gateway.streaming.*` writes the streaming block nested under a `gateway:` key in config.yaml, but the config loader only checked for a top-level `streaming:` key — silently ignoring the nested variant. Fall back to `yaml_cfg['gateway']['streaming']` when the top-level key is absent, matching the pattern already used for other nested config sections. Closes NousResearch#25676
…iled When the stream consumer's got_done handler successfully delivers the final response content via _send_or_edit but the subsequent edit (e.g. cursor removal) fails, final_response_sent remains False even though the user has already received the final answer. The gateway's fallback send path then re-delivers the same content, causing the user to see the response twice on Telegram. Introduce a new _final_content_delivered flag on the stream consumer, set by the got_done handler when the final content has reached the user. The _run_agent suppression logic now treats this flag as an additional signal (alongside final_response_sent and response_previewed) that final delivery is already complete. This preserves the existing behavior for intermediate-text-only streams (where already_sent=True but no final content has been delivered) — those still receive the gateway's fallback send, matching the test expectation in test_partial_stream_output_does_not_set_already_sent. Adds TestFinalContentDeliveredSuppression with two cases covering both the suppression (content delivered + edit failed) and the non-suppression (intermediate text only) branches.
NousResearch#25929) When codex app-server fails outside the OAuth-classified path (non-auth turn/start errors, plain TimeoutErrors, generic turn-ended status, subprocess silently exits, hard deadline timeout), the user got a bare 'Internal error' / 'turn/start failed: ...' with no context. Diagnosing config/provider/auth-bridge issues forced a re-run with verbose codex flags. Add a _format_error_with_stderr helper that appends the last few stderr lines via agent.redact.redact_sensitive_text(force=True), and use it at every catch-all error site: - ensure_started() failures (codex init / thread/start) now return a TurnResult.error with should_retire=True instead of bubbling - non-OAuth turn/start CodexAppServerError / TimeoutError - subprocess-died branch (previously dumped raw stderr_blob[-300:] with no redaction — a leak risk) - turn ended with non-completed status - hard turn-timeout deadline OAuth-classified failures and the post-tool quiet watchdog already produce clean hints and stay unchanged. The redactor catches sk-*, gh*_*, Authorization: Bearer, query-string tokens, JWTs, private keys, etc., so provider error payloads can't leak into chat output or trajectories. Inspired by openclaw#80718, adapted for our app-server transport.
…#25967) The spinner already shows tool activity visually; the 1.2 kHz tone on every tool.started event was unwanted noise (especially on WSL2, where each beep also triggers Windows Terminal's bell notification). Removed the play_beep call in _on_tool_progress entirely. Record start/stop beeps (gated by voice.beep_enabled) are unaffected.
…ize (NousResearch#25975) When the terminal shrinks, already-printed box-drawing rules (response, reasoning, streaming TTS, background-task Panels) reflow into multiple narrower rows — visible as duplicated horizontal separators / ghost lines in scrollback. Similarly, prompt_toolkit redraws a fresh status bar on SIGWINCH on top of one the terminal just reflowed, producing double-bar artifacts on column shrink. Two surgical changes: 1. Decorative scrollback boxes now use a new `HermesCLI._scrollback_box_width()` helper that clamps to `max(32, min(width, 56))`. The live TUI footer is unaffected and still uses the full width. Covers: streaming response box (open + close), reasoning box (open + close, both streaming and post-stream paths), streaming-TTS box close, final-response Rich Panel, and the background-task Rich Panel. 2. `_recover_after_resize()` now also sets a new `_status_bar_suppressed_after_resize` flag so the dynamic status bar and both input separator rules stay hidden until the next user input. The flag is cleared in the process loop the moment the user submits their next prompt, restoring chrome cleanly. Tests: - New `test_input_rules_hide_after_resize_until_next_input` covers the flag's effect on rule heights. - New `test_scrollback_box_width_caps_to_resize_safe_value` covers the helper at floor / cap / mid-range / overflow. - Existing resize-recovery test extended to assert the flag flips. Refs: NousResearch#18449 NousResearch#19280 NousResearch#22976 Salvage of NousResearch#24403. Co-authored-by: Szymonclawd <szymonclawd@mac.home>
- Treat same-dimension resize events in alt-screen mode as a repaint signal, because terminal hosts can reflow or restore the physical buffer without changing columns/rows. - Ensure pending resize erases are emitted even when the virtual diff is empty, so stale physical glyphs are still cleared. - Extract alt-screen resize repaint into prepareAltScreenResizeRepaint() for readability. - Add defensive clearTimeout in prepareAltScreenResizeRepaint so rapid resize bursts don't stack redundant delayed repaints. - Add a focused regression test for same-dimension alt-screen resize healing. Addresses NousResearch#18449 Related to NousResearch#17961
…esearch#25969) Adds 'hermes proxy start' — a local HTTP server that lets external apps (OpenViking, Karakeep, Open WebUI, ...) use a Hermes-managed provider subscription as their LLM endpoint. The proxy attaches the user's real OAuth-resolved credentials to each forwarded request, refreshing them automatically; the client can send any bearer (it gets stripped). Ships with one adapter — Nous Portal. The UpstreamAdapter ABC and registry in hermes_cli/proxy/adapters/ are designed for additional OAuth providers to plug in by name without server changes. Commands: hermes proxy start [--provider nous] [--host 127.0.0.1] [--port 8645] hermes proxy status hermes proxy providers Allowed Portal paths: /v1/chat/completions, /v1/completions, /v1/embeddings, /v1/models. Anything else returns 404 with a clear error pointing at the allowed list. aiohttp is gated like gateway/platforms/api_server.py (try-import, clean runtime error if missing). No new core dependency. Tests: 24 unit tests + 1 separate E2E that spawns the real subprocess and verifies the upstream receives the right bearer with the client's header stripped.
Adds optional channel-context backfill for Discord shared-channel sessions so the agent can see recent messages it missed between its own turns (typically when require_mention=true filters out most traffic). Previously the agent only saw the @mention message that triggered it, which led to disorienting replies in active multi-user channels where the conversation context was invisible. With backfill enabled, a configurable number of recent messages are fetched per-turn and prepended to the trigger message as a context block, kept separate from sender-prefix logic so attribution remains clean. This re-opens the work from NousResearch#13063 (approved by @OutThisLife on 2026-04-20, closed when I closed the branch to address the simpolism:main head-branch issue plus an ordering bug I caught later in live use). Filing against the freshly-rewritten problem statement in NousResearch#13054 so the design is grounded in the failure mode rather than the implementation shape. The implementation follows the **push-mode last-self-anchored** design from the two options laid out in NousResearch#13054. See the issue for the trade-off discussion vs pull-mode (NousResearch#13120 was an earlier closed PR using that shape). Treating this as a reference implementation — happy to rewrite as last-trigger anchoring or as a hybrid with NousResearch#13120 if maintainers prefer. Changes: - gateway/platforms/discord.py: - new `_discord_history_backfill()` / `_discord_history_backfill_limit()` helpers (config.extra > env > default), mirroring the existing `_discord_require_mention()` shape - new `_fetch_channel_context()` that scans `channel.history()` backwards from the trigger to the bot's last message (or limit), formats as `[Recent channel messages] / [name] msg / ...`, respects DISCORD_ALLOW_BOTS, skips system messages - per-channel `_last_self_message_id` cache to narrow the fetch window on hot paths (avoids full history scan when the bot has spoken recently) - **IMPORTANT**: passes `oldest_first=False` explicitly to `channel.history()`. discord.py 2.x silently flips the default to True when `after=` is supplied, which would select the EARLIEST N messages after our last response instead of the LATEST N before the trigger. In high-traffic windows this would return stale tool traces and drop the actual final answer the user is asking about. See regression test below. Caught in live use during a Codex tool-trace burst on May 13 2026. - gateway/config.py: discord_history_backfill + discord_history_backfill_limit settings + yaml→env bridge - gateway/platforms/base.py: channel_context field on MessageEvent - gateway/run.py: prepend channel_context after sender-prefix so the [sender name] tag applies to the trigger message alone, not to the backfill - hermes_cli/config.py: defaults for new discord.history_backfill and discord.history_backfill_limit keys - cli-config.yaml.example: documented defaults - tests/gateway/test_discord_free_response.py: 7 new tests covering cold-start backfill, self-message stop boundary, other-bot filtering, cache hot-path narrowing, stale-cache fallback, shared-channel + per-user backfill paths, and the ordering regression test (`test_fetch_channel_context_cache_uses_latest_window_when_after_set`) - tests/gateway/test_config.py: yaml→env bridge tests - tests/gateway/test_session.py: prefix-order edge cases - website/docs/user-guide/messaging/discord.md: env vars + config keys + usage docs Tested on Ubuntu 24.04 — empirically validated in my own multi-bot Discord research server for the past three weeks. Fixes NousResearch#13054 Supersedes NousResearch#13063 (closed)
Follow-up to snav's PR NousResearch#25463 contribution: flip default to on, broaden scope so backfill fires whenever require_mention gates the bot (not just shared-session channels). Why: - The mention-gate creates a session-transcript gap regardless of whether the channel is shared or per-user. In per-user sessions, Alice's session is still missing other participants' messages and her own pre-mention messages — backfill fills both gaps. - Threads naturally scope to thread-only history because discord.py's channel.history() on a thread returns only that thread's messages. - DMs still skip — every DM triggers the bot, so the session transcript is already complete. Changes: - hermes_cli/config.py: discord.history_backfill default → true - gateway/platforms/discord.py: drop the _is_shared gate, keep _is_dm skip and _needed_mention gate; env var DISCORD_HISTORY_BACKFILL default → 'true' - cli-config.yaml.example + website docs: update defaults and prose; add the DISCORD_HISTORY_BACKFILL / _LIMIT env var rows that were documented in the PR description but missing from the env-var table - tests/gateway/test_discord_free_response.py: - flip test_discord_per_user_channel_does_not_backfill → test_discord_per_user_channel_backfills_too (new behavior) - add test_discord_dm_does_not_backfill (DM skip is invariant) - give FakeThread a no-op history() so existing thread tests don't hit a fake discord.Forbidden when backfill now fires on threads too Tests: 160/160 in target files; 400/400 across all tests/gateway/ -k discord.
The prebuild step used `rm -rf` and `cp -r`, which fail on Windows (`'rm' is not recognized`). Replace with an inline Node one-liner using fs.rmSync / fs.cpSync so the build works on Windows, macOS, and Linux without adding a dependency.
…Research#25978) Pre-existing diagnostics below an edit point used to surface as 'LSP diagnostics introduced by this edit' whenever the edit deleted or inserted lines. The delta-filter key included the diagnostic's range, so the same logical error reported at a different line in the post-edit snapshot looked like a brand new diagnostic. Concrete case: deleting 14 lines in cli.py caused Pyright errors at lines 9873, 10590, 12413, 13004 (unrelated to the edit) to be reported as introduced by it. Fix: build a piecewise-linear line-shift map (via difflib's SequenceMatcher) from pre and post content, and remap baseline diagnostics into post-edit coordinates before the set-difference. Diagnostics in deleted regions drop out cleanly; diagnostics below the edit shift by the right amount; diagnostics above are untouched. The strict (range-aware) equality key stays — so a genuinely new instance of an identical error class at a different line still surfaces as new. Pieces: - agent/lsp/range_shift.py — build_line_shift, shift_diagnostic_range, shift_baseline. Pure functions, no LSP state. - agent/lsp/manager.py — LSPService.get_diagnostics_sync gains an optional line_shift kwarg; baseline is shift_baseline'd before computing the seen-set. _diag_key keeps the strict range key. - tools/file_operations.py — write_file captures pre_content for any LSP-handled extension (not just LINTERS_INPROC) and passes pre/post to _maybe_lsp_diagnostics, which builds the shift map. - New _lsp_handles_extension helper guards the pre_content read. Trade-offs preserved: - Genuinely new same-class errors at different lines still surface (content-only key would have swallowed them). - Pre-existing errors at unshifted positions still get filtered (covered by the strict-key path with no shift). - Best-effort: when pre_content can't be captured (file didn't exist, permissions), the unshifted comparison still catches most pre-existing errors; the edge case it misses is a new file with a non-empty baseline, which is structurally impossible.
Three Windows-only bugs in the web-dashboard build path. Each is small, scoped, and verified end-to-end on Windows 11 — including under a stock cmd.exe / PowerShell console with its default cp1252 encoding. 1. `sync-assets` shells out to Unix-only commands web/package.json hard-codes `rm -rf … && cp -r …`. Neither exists on Windows cmd.exe. `hermes_cli/main.py::_build_web_ui` runs npm via subprocess (which on Windows defaults to cmd.exe), so the prebuild hook crashed before Vite ever ran and the dashboard never built. Fix: web/scripts/sync-assets.mjs — ~20 lines of Node using fs.rmSync + fs.cpSync (stdlib, Node >= 16.7). No new deps, identical behavior on POSIX and Windows. 2. Build failures were silent _build_web_ui ran both subprocess calls with capture_output=True and never relayed the captured buffers on failure. Users saw 'Web UI build failed' and nothing else — no stdout, no stderr, no hint that the real problem was 'rm is not recognized'. Fix: inner _relay() helper that decodes and prints stdout + stderr (utf-8, errors='replace') whenever a step returns non-zero. Replaces the existing stderr_tail-only relay on the build path; success path is unchanged. (stderr_tail is preserved for the stale-dist fallback branch added by NousResearch#23817.) Salvaged from NousResearch#13368 by @johnisag onto current main. Conflict resolution preserves main's improvements: - _run_npm_install_deterministic() (replaces bare subprocess.run for npm install) - npm-build retry-after-sleep for Windows boot-time races (NousResearch#23817) - stale-dist fallback for non-interactive callers (NousResearch#23817) Closes NousResearch#25073, NousResearch#13368.
Codex review pointed out that even with the sync-assets fix applied,
_build_web_ui still crashes on a stock Windows console before reaching
npm: Python stdout defaults to cp1252 (or similar) and raises
UnicodeEncodeError when print() hits the arrow/check glyphs used for
status messages (→, ✗, ⚠, ✓). Reproduced locally in PowerShell:
$ PYTHONIOENCODING=cp1252 python -c "from hermes_cli.main import _build_web_ui; _build_web_ui(Path('web'), fatal=True)"
UnicodeEncodeError: 'charmap' codec can't encode character '\u2192' ...
The previous PR body claimed "end-to-end verified on Windows 11", but
that was under the venv's default (utf-8) stdout. A plain `py` or
PowerShell invocation would still fail before sync-assets ever ran.
Fix: inner _say() helper that falls back to
text.encode(sys.stdout.encoding, errors="replace")
when print() raises UnicodeEncodeError. Glyphs degrade to '?' on
ASCII / cp1252 consoles; utf-8 consoles are unaffected. Verified the
full build pipeline runs to completion with PYTHONIOENCODING=cp1252.
Scoped tightly to _build_web_ui (the function this PR already touches);
other call sites in the codebase with the same risk are out of scope.
…gnal_handler The call site at line 246 is already wrapped in try/except NotImplementedError (added in NousResearch#25969). The checker just doesn't peek at surrounding context. Mark with the suppression comment so the blocking check passes.
The 'sessions' command has been registered in the central command registry since NousResearch#20805 (May 2025) and surfaces in /help and tab-completion, but the classic CLI's process_command() never had an elif branch for it. The canonical name fell through and printed 'Unknown command: sessions'. The TUI side was wired up correctly via the SessionPicker overlay; only the legacy CLI was missing the dispatch. Adds _handle_sessions_command() which mirrors /resume's no-arg behavior inline (the CLI has no overlay primitive equivalent to the TUI picker): - /sessions and /sessions list → print the recent-sessions table - /sessions <id_or_title> → delegates to _handle_resume_command Includes regression tests covering the dispatcher wiring (the original bug) plus the three handler branches.
…-ci-unblocker-after-21012 fix(ci): stabilize shared test state after 21012
AGENT_BROWSER_CHROME_FLAGS is not read by agent-browser CLI. The correct env var is AGENT_BROWSER_ARGS, with comma-separated values. This fixes Chrome 'No usable sandbox' crash on Ubuntu 23.10+ systems where AppArmor restricts unprivileged user namespaces. The detection logic was correct but the fix used the wrong environment variable name and space-separated instead of comma-separated args.
Follow-up to the sandbox-bypass env-var fix: - Update the opt-out gate so a user-provided AGENT_BROWSER_ARGS is also respected, not just the legacy AGENT_BROWSER_CHROME_FLAGS. Previously the gate only checked the broken legacy var, so a user who pre-set AGENT_BROWSER_ARGS would still get clobbered by Hermes's auto-injection. - Document AGENT_BROWSER_ARGS in .env.example, the browser feature page, and the env var reference, with notes about the auto-injection on AppArmor-restricted systems (Ubuntu 23.10+, DGX Spark, containers). - Add Anadi Jaggia to AUTHOR_MAP.
Co-Authored-By: zxy_action <zy13377@gmail.com>
zxyaction
force-pushed
the
feat/auriko-provider
branch
from
May 15, 2026 02:19
f41b3e2 to
86b53b6
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Adds Auriko (multi-LLM inference gateway,
api.auriko.ai) as a first-class built-in provider plugin using the fast-path plugin system introduced in NousResearch#20324. Plugin auto-wires to auth, config, doctor, main dispatch, and URL-to-provider mapping with no manual edits toauth.py,runtime_provider.py, ormain.py.Type of Change
Changes Made
Plugin (auto-wired — no manual registration needed):
plugins/model-providers/auriko/__init__.py—ProviderProfileregistration (28 fallback models,AURIKO_API_KEY/AURIKO_BASE_URLenv vars, aliasauriko-ai)plugins/model-providers/auriko/plugin.yaml— plugin manifestManual registrations (not yet auto-wired by plugin system):
hermes_cli/models.py—_PROVIDER_MODELS["auriko"](28-model catalog),_PROVIDER_ALIASES["auriko-ai"]hermes_cli/providers.py—HERMES_OVERLAYS,ALIASES,_LABEL_OVERRIDESentriesagent/model_metadata.py—_PROVIDER_PREFIXES(canonical + alias)Tests:
tests/hermes_cli/test_auriko_provider.py— 31 tests across 8 classes (registration, auto-wiring, model catalog, prefixes, model fetch, auxiliary client, CLI flow, providers module)tests/providers/test_plugin_discovery.py— count bump: 28→29 plugin dirs, 33→34 profilesDocs (6 files):
website/docs/integrations/providers.md— provider row + supported providers listwebsite/docs/reference/environment-variables.md—AURIKO_API_KEY/AURIKO_BASE_URL+HERMES_INFERENCE_PROVIDERlistwebsite/docs/getting-started/quickstart.md— provider rowwebsite/docs/reference/cli-commands.md—aurikoin--providerlistwebsite/docs/user-guide/features/fallback-providers.md— provider rowwebsite/docs/user-guide/configuration.md—aurikoin provider listHow to Test
scripts/run_tests.sh tests/hermes_cli/test_auriko_provider.py -v— 31 Auriko-specific testsscripts/run_tests.sh tests/providers/test_plugin_discovery.py -v— plugin discovery countsscripts/run_tests.sh— full suite regression checkhermes chat -q "hello" --provider auriko -Q(requiresAURIKO_API_KEY)hermes doctor(Auriko connectivity check)Test Results
All 55 failures are pre-existing upstream issues (file_read_guards, file_staleness, systemctl, skin integration, etc.) — none related to Auriko changes.
Checklist
Code
feat(providers):)pytest tests/ -qand all tests passDocumentation & Housekeeping
website/docs/)cli-config.yaml.exampleif I added/changed config keys — N/A (plugin auto-wires config)