feat(messaging): add explicit active handoff primitives

[verybigdog]

Summary

• add opt-in send_message(trigger_agent=True) active handoff result plumbing while keeping passive sends as the default
• add Kanban notify-subscribe --trigger-agent persistence and gateway notifier support
• actively wake the originating gateway agent for gateway-created Kanban task terminal ACKs
• preserve Kanban origin ACK edges for CLI/script-created graphs:
  • child tasks inherit parent/root terminal notification subscriptions
  • explicit Origin/return_to: Discord ... (<#channel_id>) body prose infers a notify subscription for CLI-created cards
• document that headless Kanban workers should write board verdicts instead of directly sending Discord/Telegram final ACKs

Safety

• default behavior remains passive unless trigger_agent=True / --trigger-agent is explicitly requested or a CLI-created task carries a narrow, machine-readable Discord Origin/return_to line
• active wake is best-effort after a successful visible platform send; it is not an atomic send+wake transaction
• visible platform send and active wake are reported separately via triggered_agent / trigger_error
• Kanban subscriptions wake agents only when they carry trigger_agent=1
• active wake failures are isolated from passive delivery so the notifier does not crash or retry the same terminal event just because wake is unavailable
• the Origin/return_to parser intentionally recognizes only explicit Discord channel-id shapes such as <#123...> or chat_id=123...; it is not a general natural-language parser

Tests

• python -m pytest tests/hermes_cli/test_kanban_notify.py tests/gateway/test_kanban_notifier.py tests/gateway/test_kanban_notifier_watcher_dispatch_gate.py tests/tools/test_send_message_tool.py -q → 160 passed, 2 warnings
• git diff --check
• python -m compileall gateway/kanban_watchers.py gateway/slash_commands.py hermes_cli/kanban.py hermes_cli/kanban_db.py tools/send_message_tool.py tests/hermes_cli/test_kanban_notify.py -q

Refresh notes

• rebased onto current origin/main (4829f8d2c at refresh time)
• force-pushed with --force-with-lease from old head f3c9a6453 to new head 9829f4ee0

Local operator policy and AgentFlow supervisor hooks remain intentionally out of scope.

[verybigdog]

Marked ready for review.

This PR intentionally keeps the primitive small/default-off:

• normal sends and Kanban notifications remain passive by default
• active wake is explicit (trigger_agent=True / notify-subscribe --trigger-agent)
• visible platform delivery and active wake are reported separately (triggered_agent / trigger_error)
• active wake is best-effort after a successful visible send, not an atomic send+wake transaction

One implementation detail I'm happy to adjust: _trigger_gateway_agent currently lives near send_message_tool because send_message is the first direct caller. If maintainers prefer, I can move it into a gateway-owned handoff/helper module so the Kanban notifier does not import from tools.

[verybigdog]

Refreshed this PR onto current origin/main and removed local-only scope.

Current branch contents are only the generic active handoff primitive:

• explicit trigger_agent=True send_message wake path
• visible delivery and active wake reported separately
• Kanban notify active-wake opt-in
• passive send/notify behavior remains default

Intentionally excluded from this upstream PR:

• AgentFlow gateway supervisor / external queue dispatch
• resolve-fanin command (split out separately in feat(kanban): add resolve-fanin command for review verdicts #40465)

Verification after refresh:

• git diff --check origin/main..HEAD
• python3 -m py_compile gateway/run.py hermes_cli/kanban.py hermes_cli/kanban_db.py tools/send_message_tool.py
• python3 -m pytest tests/hermes_cli/test_kanban_notify.py tests/tools/test_send_message_tool.py -q -o 'addopts=' → 148 passed

[verybigdog]

Refreshed this PR onto current origin/main and added the ACK-edge regression fix that came out of live Kanban routing.

Changes since the previous head:

• rebased branch onto origin/main (4829f8d2c at refresh time)
• added commit 9829f4ee0 fix(kanban): preserve origin ACK edges for CLI-created graphs
• child tasks now inherit parent/root terminal notify subscriptions, so child BLOCK/DONE states keep the same return path as the graph root
• CLI/script-created cards with explicit Origin/return_to: Discord ... (<#channel_id>) prose now infer a notify subscription and set trigger_agent=True
• kept AgentFlow/local operator policy out of this PR

Verification on the refreshed branch:

python -m pytest \
  tests/hermes_cli/test_kanban_notify.py \
  tests/gateway/test_kanban_notifier.py \
  tests/gateway/test_kanban_notifier_watcher_dispatch_gate.py \
  tests/tools/test_send_message_tool.py \
  -q

160 passed, 2 warnings in 4.43s

git diff --check
python -m compileall \
  gateway/kanban_watchers.py \
  gateway/slash_commands.py \
  hermes_cli/kanban.py \
  hermes_cli/kanban_db.py \
  tools/send_message_tool.py \
  tests/hermes_cli/test_kanban_notify.py \
  -q

New head: 9829f4ee03cf30b79e37b7d87c92bf7e7618158b.

[verybigdog]

Refreshed against latest origin/main and re-verified.

Evidence:

• old fork head: 9829f4ee03cf30b79e37b7d87c92bf7e7618158b
• new fork head: 3e792866f
• base: origin/main 955fa4006
• verification: python -m compileall -q tools/send_message_tool.py gateway/kanban_watchers.py gateway/slash_commands.py hermes_cli/kanban.py hermes_cli/kanban_db.py
• verification: python -m pytest tests/tools/test_send_message_tool.py tests/hermes_cli/test_kanban_notify.py tests/gateway/test_kanban_notifier.py -q => 158 passed, 2 warnings
• independent post-remediation review: GO / no blocking security or logic findings

Scope note from review hardening:

• send_message no longer exposes or honors model-supplied trigger_agent; active wake remains private to verified Kanban notification subscriptions.
• Origin/return_to body inference now creates passive terminal ACK subscriptions only. Active wake requires explicit control-plane intent via --ack-trigger-agent / notify-subscribe --trigger-agent or inherited verified subscriptions.
• ContextOps/local dogfood hooks and lane policy are intentionally excluded.

[verybigdog]

Refreshed again against current origin/main after upstream advanced.

Evidence:

• old fork head: 3e792866f90b393c01c0f5f8adf6f89b4658470c
• new fork head: e8714861d
• current base: origin/main c6b0eb4de
• conflict resolved in tools/send_message_tool.py by preserving upstream react/unreact schema and keeping trigger_agent out of the model-exposed send_message schema.
• verification: python -m compileall -q tools/send_message_tool.py gateway/kanban_watchers.py gateway/slash_commands.py hermes_cli/kanban.py hermes_cli/kanban_db.py
• verification: python -m pytest tests/tools/test_send_message_tool.py tests/hermes_cli/test_kanban_notify.py tests/gateway/test_kanban_notifier.py -q => 158 passed, 2 warnings

[nikitaBarkov]

Hi @verybigdog 👋

1. Is this PR still active and planned for merge? It currently has conflicts with main, so I wanted to check its status before making any changes on my side.

2. I need essentially the same functionality, but with a few extensions on top of your work:

• a delivery_mode enum (notify / notify+wake / wake) instead of the on/off trigger_agent, adding a pure wake mode (wake the agent without a passive message);
• a persisted chat_type, so the woken turn resolves the operator's actual session via build_session_key instead of creating a context-less parallel session;
• keeping the wake primitive out of the model-exposed send_message schema (internal gateway/CLI/notifier only);
• retaining the background wake task so it can't be garbage-collected and silently dropped.

3. Since we'd be touching many of the same files, I'd rather coordinate than open a competing PR. A couple of options, whichever you prefer:

• I open my own branch/PR and credit you as a co-author (Co-authored-by:), effectively superseding this one; or
• I fold my changes into your branch so it ships as a single feature, with you as the primary author.

Let me know what works best for you.

If useful, here's the diff of my version: main...nikitaBarkov:nikita.barkov/wake_notify_kanban

[verybigdog]

Thanks @nikitaBarkov — yes, this is still active, and I agree it needs a refresh against current main before it can realistically move forward.

Your direction looks like the right abstraction to me. In particular:

• delivery_mode (notify / notify+wake / wake) is cleaner than the old boolean-style trigger_agent gate;
• persisting chat_type so the wake resolves through build_session_key fixes the context-less parallel-session issue;
• keeping the wake primitive out of the model-exposed send_message schema is the right safety boundary;
• retaining the background wake task is important so wake delivery cannot be silently GC-dropped.

Since both branches touch the same gateway/Kanban/CLI surface and both are stale/conflicting against current main, I think the safest path is a refreshed superseding branch from current main that folds in these semantics, with you credited as co-author for the design/diff. I’ll avoid opening a competing implementation that ignores your work.

I’ll use this as the target shape for the refresh:

• delivery_mode enum persisted on Kanban notify subscriptions;
• persisted chat_type for wake session-key correctness;
• internal-only wake primitive, not model-exposed through send_message;
• retained background wake task;
• passive delivery and active wake reported separately;
• fail-closed tests around notify/wake behavior and session routing.

If you strongly prefer this to land through your branch instead, say so and I can adapt, but a clean current-main branch will probably be easier to review.

[verybigdog]

Refreshed this PR on current origin/main with Nikita Barkov's delivery_mode / chat_type active-wake design folded in and credited in the commit trailer.

Head: 1f5101efd8ea805e0349bccf1560302724d9fffb

What changed:

• kanban_notify_subs now persists delivery_mode (notify, notify+wake, wake) and chat_type.
• Plain/default notify remains passive and never wakes, including legacy rows with a session_id.
• notify+wake and wake use an internal-only _trigger_gateway_agent helper; no trigger_agent / wake field is exposed through the model-facing send_message schema.
• Wakes replay persisted chat_type, user_id, and notifier profile routing so build_session_key lands in the real operator session.
• The scheduled wake task is retained in the gateway background task set.
• Status/dashboard churn is passive only; active wake fires for terminal outcome kinds.

Verification run locally from the refreshed worktree:

• scripts/run_tests.sh tests/hermes_cli/test_kanban_notify.py tests/tools/test_send_message_tool.py -> 176 passed, 0 failed.
• git diff --cached --check -> clean.
• python3 -m py_compile gateway/kanban_watchers.py gateway/slash_commands.py hermes_cli/kanban.py hermes_cli/kanban_db.py tools/send_message_tool.py tests/hermes_cli/test_kanban_notify.py tests/tools/test_send_message_tool.py -> clean.
• Independent Claude review after fix -> PASS.

PR is no longer conflicting: mergeable=MERGEABLE (GitHub currently reports mergeStateStatus=BLOCKED, likely because branch protection/checks have not completed or are required externally).

[verybigdog]

Follow-up for review BLOCK on gateway auto-subscribe delivery_mode/chat_type:

Fixed in new head dafc0e4b38aa1f660dcc4381bd24ac50f7574530.

Changes:

• gateway/session_context.py now carries HERMES_SESSION_CHAT_TYPE through the task-local session context/env bridge.
• gateway/run.py binds context.source.chat_type into that session context.
• tools/kanban_tools.py::_maybe_auto_subscribe now writes gateway-created task subscriptions with the source chat_type and delivery_mode='notify+wake'.
• TUI/desktop fallback remains passive/default (delivery_mode='notify', chat_type='dm'), and CLI/no-channel paths still do not auto-subscribe.
• Regression assertions added in tests/tools/test_kanban_tools.py.

Verification:

• scripts/run_tests.sh tests/tools/test_kanban_tools.py tests/hermes_cli/test_kanban_notify.py tests/tools/test_send_message_tool.py => 273 passed.
• python3 -m py_compile gateway/session_context.py gateway/run.py tools/kanban_tools.py tests/tools/test_kanban_tools.py => passed.
• git diff --check => passed.
• Manual contextvar adversarial probe: Discord/group tool-created child row recorded delivery_mode='notify+wake', chat_type='group', thread/user/profile preserved.

Credit/history: this is a narrow follow-up commit on top of the existing PR branch; Nikita’s existing contribution remains intact.

[verybigdog]

Final fan-in after the gateway auto-subscribe remediation:

• Current PR head verified on GitHub: dafc0e4b38aa1f660dcc4381bd24ac50f7574530 (contrib/active-handoff-primitives -> main), PR open, mergeable=MERGEABLE, mergeStateStatus=BLOCKED, no status checks currently reported by GitHub.
• Remediation review verdict: GO. The follow-up commit carries gateway-created kanban_create auto-subscriptions through delivery_mode='notify+wake' and persisted chat_type, while CLI/no-channel and TUI/desktop fallback paths remain passive/default.
• Local re-verification from a clean review worktree at the same head:
  • git diff --check origin/main...HEAD -> passed.
  • python3 -m py_compile gateway/session_context.py gateway/run.py gateway/kanban_watchers.py gateway/slash_commands.py hermes_cli/kanban.py hermes_cli/kanban_db.py tools/kanban_tools.py tools/send_message_tool.py tests/tools/test_kanban_tools.py tests/hermes_cli/test_kanban_notify.py tests/tools/test_send_message_tool.py -> passed.
  • scripts/run_tests.sh tests/tools/test_kanban_tools.py tests/hermes_cli/test_kanban_notify.py tests/tools/test_send_message_tool.py -> 273 passed, 0 failed.
  • send_message schema probe -> exposed properties are only action, emoji, message, message_id, target; no model-exposed trigger_agent, wake, delivery_mode, or chat_type field.
• Scope check: diff is limited to gateway/Kanban notify delivery-mode/session-context plumbing, tests, and Kanban docs; main workspace only has pre-existing untracked .kanban-t_4623baae/ review artifacts.

Credit/coordination: Nikita Barkov's design/diff remains credited in commit 1f5101efd8 via Co-authored-by: Nikita Barkov <nikita.barkov@users.noreply.github.com>, and this refresh intentionally treats Nikita as a co-worker/co-designer on the delivery_mode + chat_type shape rather than a competing implementation.

[teknium1]

Thanks for carrying the delivery-mode and persisted-routing design through the notifier path. The current-main premise is real: gateway/kanban_watchers.py:517-548 still hardcodes wake sources to chat_type="group".

Problems

• tools/send_message_tool.py:534-535 falls back to runner.adapters after _authorization_adapter() returns None. That reintroduces the cross-profile wrong-bot delivery that current main deliberately fixed in b225b30d0; _authorization_adapter fails closed for unavailable stamped profiles at gateway/authz_mixin.py:50-57.
• The routing data is still incomplete. The PR persists user_id, but build_session_key prefers user_id_alt for group/channel isolation (gateway/session.py:936-957). A source carrying a distinct alternate identifier can still wake a parallel session.

Suggested changes

• Preserve the profile chokepoint: return a trigger error rather than falling back to the default adapter for an unavailable non-default profile, with a regression test.
• Persist and replay user_id_alt as well, and test equality between the original and woken build_session_key when the two identifiers differ.

Automated hermes-sweeper review.

[teknium1]

Blocking: do not fall back to runner.adapters after _authorization_adapter(platform, profile) returns None for a stamped secondary profile. Current main deliberately fails closed here (gateway/authz_mixin.py:50-57; b225b30d0) because this fallback sends via the default profile's bot. Return a trigger error for that case and add the same no-default-fallback regression.

[teknium1]

Blocking: this routing row records user_id but not user_id_alt. build_session_key() prefers user_id_alt for non-DM participant isolation (gateway/session.py:936-957), so the wake can still create a parallel session. Persist and replay the alternate identifier, with a differing-ID session-key regression.

[verybigdog]

Refreshed #37865 on current origin/main and addressed the 2026-07-13 sweeper findings.

Head: e5718e67f6bddc632db7b5d35a1cfc3c50b80f52 (rebased on origin/main 569b912d7d0931c7256e9f5fb326609e9deda377).

Fixes in this refresh:

• Active wake now preserves the profile authorization chokepoint: when _authorization_adapter() cannot resolve a stamped non-default profile adapter, _trigger_gateway_agent() returns trigger_error and does not fall back to the default/wrong bot adapter. Regression: test_profile_stamped_wake_fails_closed_without_profile_adapter.
• kanban_notify_subs now persists/inherits user_id_alt; gateway-created auto-subscriptions capture HERMES_SESSION_USER_ID_ALT; notifier wake replay passes user_id_alt into SessionSource.
• Routing coverage now proves the original and woken build_session_key stay equal when user_id and user_id_alt differ for DM, isolated group, isolated channel, shared group, and thread cases.
• Preserved the existing internal-only wake boundary: send_message model schema still exposes only action, emoji, message, message_id, target; no trigger_agent, wake, delivery_mode, chat_type, or user_id_alt field.
• Preserved retained background wake tasks, passive default delivery, and separate visible-delivery / active-wake receipts.
• Nikita Barkov's credited delivery_mode/chat_type design commit remains in history with its Co-authored-by: trailer.

Verification:

• git diff --check origin/main...HEAD -> passed.
• python3 -m py_compile gateway/session_context.py gateway/run.py gateway/kanban_watchers.py gateway/slash_commands.py hermes_cli/kanban.py hermes_cli/kanban_db.py tools/kanban_tools.py tools/send_message_tool.py tests/tools/test_send_message_tool.py tests/tools/test_kanban_tools.py tests/hermes_cli/test_kanban_notify.py -> passed.
• Focused wrapper: scripts/run_tests.sh tests/tools/test_send_message_tool.py tests/hermes_cli/test_kanban_notify.py tests/tools/test_kanban_tools.py tests/gateway/test_kanban_notifier.py tests/gateway/test_kanban_notifier_watcher_dispatch_gate.py tests/gateway/test_session_env.py -q -> 305 passed, 0 failed.
• Full wrapper: scripts/run_tests.sh -q -> 41,234 passed, 16 failed. The failures are in existing unrelated areas not touched by this PR diff (tests/agent/test_codex_responses_adapter.py, tests/agent/test_skill_utils.py, tests/gateway/test_agent_cache.py, tests/hermes_cli/test_gateway_runtime_health.py, tests/honcho_plugin/test_pin_peer_name.py, tests/tools/test_execute_code_approval_cluster.py, tests/tools/test_skills_tool_discovery_cache.py).
• GitHub after push: mergeable=MERGEABLE, mergeStateStatus=BLOCKED, no status checks currently reported.

[verybigdog]

Verdict: BLOCK

I re-reviewed current head e5718e67f6bddc632db7b5d35a1cfc3c50b80f52 against origin/main 569b912d7d0931c7256e9f5fb326609e9deda377.

Blocking finding:

• gateway/slash_commands.py:466-480 still drops source.user_id_alt when /kanban create auto-subscribes a gateway-created task. The row is created with delivery_mode='notify+wake', but only user_id and chat_type are persisted. For isolated group/channel sessions where build_session_key() prefers user_id_alt, the later active wake replays a different source and lands in a parallel context-less session.

Repro probe on this PR checkout:

original           agent:main:discord:group:chat-1:union-id
replay_missing_alt agent:main:discord:group:chat-1:open-id
equal              False

The model-tool auto-subscribe path in tools/kanban_tools.py does persist HERMES_SESSION_USER_ID_ALT, but the gateway slash-command auto-subscribe path does not, so the fail-closed/routing fix is incomplete.

Suggested fix:

• In gateway/slash_commands.py, read source.user_id_alt and pass user_id_alt=... into kb.add_notify_sub(...) alongside user_id.
• Add a regression test for /kanban create gateway auto-subscribe with user_id != user_id_alt, asserting the persisted subscription includes user_id_alt and the original/wake build_session_key() are equal.

Checks run from clean worktree /tmp/hermes-pr37865-review:

• git diff --check origin/main...HEAD -> pass
• python3 -m py_compile gateway/session_context.py gateway/run.py gateway/kanban_watchers.py gateway/slash_commands.py hermes_cli/kanban.py hermes_cli/kanban_db.py tools/kanban_tools.py tools/send_message_tool.py tests/tools/test_send_message_tool.py tests/tools/test_kanban_tools.py tests/hermes_cli/test_kanban_notify.py -> pass
• scripts/run_tests.sh tests/tools/test_send_message_tool.py tests/hermes_cli/test_kanban_notify.py tests/tools/test_kanban_tools.py tests/gateway/test_session_env.py -q -> 295 passed

Because the current PR acceptance criteria explicitly require user_id_alt persistence/replay and exact session-key equality across group/thread variants, this remains BLOCK until the gateway slash auto-subscribe path is covered too.

[verybigdog]

Remediation for reviewer BLOCK #37865 (comment) pushed at a6ee9a2.

What changed:

• gateway /kanban create auto-subscribe now persists source.user_id_alt beside user_id, chat_type, thread_id, notifier_profile, and delivery_mode="notify+wake".
• Added a real slash-command-path regression that creates the subscription through GatewayRunner._handle_kanban_command, reloads the persisted row, reconstructs the wake SessionSource, and proves build_session_key(original) == build_session_key(replayed) for user_id != user_id_alt in channel and thread-isolated shapes.

Verification:

• scripts/run_tests.sh tests/hermes_cli/test_kanban_notify.py → 24 passed.
• python3 -m py_compile gateway/slash_commands.py tests/hermes_cli/test_kanban_notify.py → passed.
• git diff --check → passed.

Live PR state after push: mergeable=MERGEABLE, mergeStateStatus=BLOCKED, checks=[] (no checks reported yet).

[verybigdog]

Keeping this as the canonical active-wake PR for the M28 consolidation. HEAD (a6ee9a2) already resolves the sweeper review: it fails closed for an unavailable non-default stamped profile adapter (no default-bot fallback) and persists+replays user_id_alt so a woken turn keys to the same build_session_key as the originating event. delivery_mode/chat_type/user_id_alt, parent->child notify-sub inheritance (including delivery_mode/chat_type/user_id_alt), DM/group/thread routing, and 'no model-exposed wake / no body-prose inference' all land here. Targeted tests: 281 passed via scripts/run_tests.sh; merges cleanly against current main. The distinct follow-on surface is delivery-boundary failure receipts (#49752); #49751 is duplicate because #37865 already carries the task-graph subscription inheritance.

[AlexanderPrendota]

@teknium1 can u have a look please? :)

[AlexanderPrendota]

@verybigdog, hey, could u please resolve the conflicts?

[GottZ]

This was generated by AI during triage.

Summary

Four PRs address the active-wake and durable Kanban ACK path: #37865 implements opt-in delivery modes with source-faithful wake routing and graph subscription inheritance; #49593 adds receipt/ledger and diagnostics machinery around an overlapping wake implementation; #49749 isolates wake-text sanitization; and #49751 isolates graph subscription inheritance plus origin wiring.

Related pull requests

• feat(messaging): add explicit active handoff primitives #37865 related — (+1223/-114) — n/a: Implements passive-by-default notify/notify+wake/wake delivery, persists and replays chat_type, user_id, user_id_alt, and profile ownership, and carries subscriptions into child tasks. The diff addresses the contributor keep_open review's wrong-profile and parallel-session findings by failing closed for unavailable stamped profiles and round-tripping user_id_alt; the PR remains relevant as the open canonical implementation but currently needs conflict resolution.
• feat(gateway): add active-wake operator receipts #49593 [closed] related — (+3249/-25) — n/a: Adds scheduled/accepted/started wake receipts, an ACK ledger, relay diagnostics, and an overlapping active-wake path. Despite the contributor keep_open review, this closed PR should remain a reference rather than be reopened: its diff hard-codes chat_type="group" and records operator observed on session acceptance/start, while feat(messaging): add explicit active handoff primitives #37865 provides source-faithful routing; any independently useful receipt or diagnostic pieces require a narrow follow-up that preserves the review's operator-observation distinction.
• fix(send_message): sanitize active-wake snippets #49749 [closed] duplicate — (+561/-2) — n/a: Adds redaction for token-, path-, and media-like content but also duplicates the active-wake transport with a model-facing trigger and a synthetic group participant. Despite the contributor keep_open review, it is appropriately closed as superseded by feat(messaging): add explicit active handoff primitives #37865 because the diff neither truncates the wake text as claimed nor routes to the originating session; sanitization alone can be salvaged separately if still needed.
• fix(kanban): preserve ACK subscriptions across task graphs #49751 [closed] duplicate — (+279/-17) — n/a: Copies parent notification subscriptions into child tasks and adds explicit origin options, but persists an unconsumed trigger_agent flag and infers external targets from task-body prose. Despite the contributor keep_open review, it is appropriately closed as a duplicate of feat(messaging): add explicit active handoff primitives #37865, whose diff preserves the same graph edge together with the routing fields and watcher consumption that this PR lacks.

Duplicates

#49593 and #49749 substantially duplicate #37865's active-wake mechanism; #49751 duplicates #37865's Kanban graph-subscription inheritance. #49593 also incorporates the narrower surfaces represented by #49749 and #49751.

Suggested consolidation

Author action on #37865: rebase onto current main and resolve the reported conflicts, then retain its source-faithful, profile-fail-closed delivery-mode and graph-inheritance path while addressing any newly exposed integration failures. Keep #49593, #49749, and #49751 closed as superseded/duplicate implementations; salvage only narrow receipt/diagnostic or sanitization pieces in separate follow-ups that preserve the contributor reviews' routing and operator-observation constraints.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    subgraph Dup37865 ["PRs duplicating each other"]
        P37865["PR #37865 (open)"]
        P49749["PR #49749 (closed)"]
        P49751["PR #49751 (closed)"]
    end
    class P37865 open
    class P49749 closed
    class P49751 closed
    class P37865 target
    click P37865 "https://github.com/NousResearch/hermes-agent/pull/37865"
    click P49749 "https://github.com/NousResearch/hermes-agent/pull/49749"
    click P49751 "https://github.com/NousResearch/hermes-agent/pull/49751"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 4 pull requests and 0 issues in this complex. Each diff was read against this issue; Assessment working set: 279 kB of PR diffs, 5 kB of issue/PR text, 18 kB of discussion (23 comments), 0 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.

[verybigdog]

Resolved the conflicts against current main (3dcbe9001f30de749971911041c02916437b5bff) in 224a85b5ca1cfabc7c73fa93c430ca2b1988a30e and pushed the updated branch.

The resolution keeps the PR's explicit notify / notify+wake / wake modes, source-faithful chat_type / user_id_alt routing, and inherited return-path metadata, while using the current gateway.wake.deliver_wake path rather than restoring the older duplicate wake implementation.

Validation:

• PR is now reported MERGEABLE by GitHub.
• Targeted notifier/routing/session tests: 112 passed.
• ruff check on changed Python files: passed.
• Full local suite was also attempted; it was not clean in this checkout because optional test dependencies (acp, Daytona, Modal, fal-client, etc.) are absent and the run timed out at 87%. The PR-targeted suite had no failures.

[teknium1]

Merged via PR #85487 with your work applied onto current main. Your original commits were authored under a local placeholder identity (hermes-agent@users.noreply.local), so the salvage carries the change as a single commit re-attributed to your public email — credit preserved in git log and the PR body.

We added two follow-up fixes on top: (1) the migration now backfills pre-existing gateway subscriptions to notify+wake on first-add of the column (before delivery_mode existed, wake was unconditional — defaulting legacy rows to notify would have silently disabled it on upgrade), and (2) api_server subscriptions default to notify+wake since that platform has no push send() and the wake self-post is the delivery.

Thanks for the design and the thorough test coverage — the delivery-mode gate and the user_id_alt round-trip were the right call.