The Web UI disables the message input for any conversation whose
platform_type is not 'web' because Scope B (bidirectional bridging) isn't
shipped yet. The old disabledReason string — "Continuing chats from other
platforms in the Web UI is coming soon" — was both vague and increasingly
misleading now that Telegram conversations render their full history in
the Web UI.

Replace the hardcoded string with a platform keyed lookup map so each
platform gets a clear "reply from the originating app" hint. The disable
condition itself is unchanged; this is pure copy + a small constant.

Only telegram is functionally wired up (persistence hooks land in a
sibling commit); slack/discord/github entries are forward-compatible and
take effect as soon as persistence is broadened to those platforms.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add a focused test suite for the telegram persistence gate added to
handleMessage. Covers the three load-bearing cases on the user-message
side of the gate:

1. Natural-language telegram messages persist exactly one row with role
   'user', the DB conversation id, the raw message text, and metadata
   { platformType: 'telegram' }.
2. Deterministic slash commands (/help) skip persistence entirely —
   neither user nor assistant rows are created.
3. Web-platform conversations do NOT trigger the centralized path, so
   web's existing MessagePersistence buffer still owns that flow.

Assistant-message persistence hooks (inside handleStreamMode,
handleBatchMode, and the top-level catch) require mocking sendQuery to
yield actual content, which needs a more elaborate mock setup than the
existing test file provides. Tracking that as a follow-up rather than
blocking the MVP on it — the user-persistence path is the primary new
logic and is covered here.

A new mock.module('../db/messages', ...) is added near the existing DB
mocks so that orchestrator-agent.ts's new messageDb import does not try
to open a real DB connection. orchestrator-agent.test.ts runs in its
own bun test invocation per packages/core/package.json, so the new
mock does not pollute sibling orchestrator tests.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

This repo is a clone of coleam00/Archon and will evolve upstream as the
project moves through beta. To keep local customizations persistent
across upstream releases without merge chaos, the working copy is set up
fork-first: origin → matzls/Archon (push access), upstream →
coleam00/Archon (read-only).

Add a "Fork & Upstream Integration" section to CLAUDE.md next to the
existing Git Workflow content so future sessions have a single grounded
reference for:

- Which remote is which and what dev tracks
- Where different customization types belong (personal config vs.
  upstreamable code changes vs. personal code changes)
- The exact commands to integrate upstream releases (fetch + ff merge +
  rebase feature branches)
- The exact commands to contribute back via gh pr create

Intentionally short — this is routing guidance, not a git tutorial.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

9-node DAG PIV loop tuned for Codex behavioral tendencies:
numbered SIGNAL EMISSION CONTRACTs, task-scoped implement loop
(no repo-wide validators mid-task), pre-existing failure tolerance
in code-review, per-file git staging, and tightened COMPLETE signal.

Validated end-to-end on my-second-brain-build (Python/Obsidian vault)
with 32 pre-existing ruff violations — workflow correctly scoped fixes
to branch-introduced issues only.

Also add root-level artifacts/ to .gitignore (workflow runtime output).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Reuse the deterministic slash-command allowlist for Telegram user-message persistence so slash-prefixed AI prompts are stored while ephemeral commands still skip persistence.

Add a regression test covering /etc/hosts and stabilize the command-parser mock in the Telegram persistence test block.

Co-authored-by: Codex <noreply@openai.com>

Context: preserve and land the Codex-specific assist workflow onto current dev while keeping the newer telegram persistence behavior already present on dev.

Change:
- add the bundled archon-assist-codex command and workflow defaults plus the tracked Archon skill files they depend on
- default continue and orchestrator assist routing to archon-assist-codex when the assistant type is codex
- extend server, web, docs, and core test coverage for the new workflow and the assistant-aware prompt-builder signatures

Validation:
- bun test packages/cli/src/commands/continue.test.ts
- bun test packages/core/src/orchestrator/prompt-builder.test.ts
- bun test packages/core/src/orchestrator/orchestrator.test.ts
- bun test packages/server/src/routes/api.health.test.ts
- bun test packages/server/src/routes/api.workflows.test.ts
- bun test packages/web/src/lib/workflow-metadata.test.ts
- bun test packages/workflows/src/defaults/bundled-defaults.test.ts
- bun --filter @archon/cli type-check
- bun --filter @archon/core type-check
- bun --filter @archon/server type-check
- bun --filter @archon/workflows type-check
- bun --filter @archon/web type-check
- bun run validate

Codex-Session: 019d80c8-3cb7-79b1-8443-d09a42cb5020
Codex-Rollout: sessions/2026/04/12/rollout-2026-04-12T10-21-39-019d80c8-3cb7-79b1-8443-d09a42cb5020.jsonl
Co-authored-by: Codex <noreply@openai.com>

Extract the detect-project workflow node into a reusable Bun script while preserving its stdout contract. Also tighten the Codex loop prompts so feedback fixes use per-file staging, scoped validation stays tool-accurate, and iteration violations fail without rewriting history.

Extend typed ESLint coverage to .archon/scripts so the new script participates in the existing pre-commit checks.

Co-authored-by: Codex <noreply@openai.com>

Document how assistant selection works across host skills, conversations, workflows, and nodes. Capture fork-specific Codex additions, upstream differences, and the current Codex limitations for workflow nodes.

Co-authored-by: Codex <noreply@openai.com>

Preview upstream sync onto custom dev; auto-merged cleanly and passed bun run validate.

Co-authored-by: Codex <noreply@openai.com>

Document the phase-by-phase session model for archon-piv-loop-codex,
including fresh-context boundaries, interactive loop resume behavior,
and a flow diagram for future review.

Co-authored-by: Codex <noreply@openai.com>

Promote archon-piv-loop-codex into the bundled default workflow set and update the overview docs to advertise it there.

Also fix the execution-notes path references after the workflow moved under .archon/workflows/defaults.

Co-authored-by: Codex <noreply@openai.com>

Exclude CLI-backed workflow runs from the server startup orphan-failure sweep and document the distinction in the workflow authoring guide.

This keeps server restarts from marking active CLI executions failed while they continue in a separate process against the same database.

Co-authored-by: Codex <noreply@openai.com>

Add a reusable reference for repeated Archon log-debugging sessions and surface it from the Codex assist lane and the top-level Archon skill routing table.

The new guide explains the three log layers, run discovery, JSONL filtering, event interpretation, and when to use UI or raw logs.

Co-authored-by: Codex <noreply@openai.com>

- bundle the detect-project helper as a default script and resolve Archon default scripts when repo-local scripts are absent
- stop PIV loop nodes early when git HEAD and task-progress tracking stop advancing
- fail workflow CLI commands early when ~/.archon is not writable and clarify the sandbox failure mode in docs
- persist richer DAG failure metadata for partial-run diagnostics

Co-authored-by: Codex <noreply@openai.com>

- add routing guidance for monitoring, interactive relays, and log debugging in the Archon skill
- add focused references for workflow monitoring cadence, paused-run relay behavior, and JSONL-first debugging
- keep Archon follow-up handling grounded in the run status and per-run logs

Co-authored-by: Codex <noreply@openai.com>

Add the generated PRD for workflow node display names to the Archon repo under docs/prd.

The document keeps one PRD with a small execution-graph-only phase 1 and defers builder, non-graph execution surfaces, inference, and historical-fidelity questions to phase 2.

Co-authored-by: Codex <noreply@openai.com>

Add the fork-level design doc that defines the Codex-first workflow surface, decision rules, and follow-on implementation sequence.

Co-authored-by: Codex <noreply@openai.com>

Refine the Archon Codex skill and assist command so substantial implementation work routes to the Codex PIV lane, and add explicit worktree-proof/readback guardrails for assist-mode edits.

Co-authored-by: Codex <noreply@openai.com>

Add a full Codex-first operator and authoring surface for the Archon skill, including workflow monitoring, debugging, repo init, command authoring, DAG authoring, CLI references, configuration guidance, and a Codex capability crosswalk.

Correct the documented Codex parity boundaries so loop model/provider overrides are described accurately and workflow-level Codex tuning fields are called out as parsed but not runtime-effective per workflow in the current executor.

Validation:
- git diff --check
- archon workflow list --json

Co-authored-by: Codex <noreply@openai.com>

Make modelReasoningEffort, webSearchMode, and additionalDirectories effective from workflow YAML for Codex execution, with config fallback for normal and loop nodes.

Add regression coverage for override, fallback, and mixed-provider loop preservation. Update Codex Archon references to match the implemented precedence.

Co-authored-by: Codex <noreply@openai.com>

Update the assistant architecture reference to reflect that workflow-level Codex tuning fields now override Archon config with config fallback, matching the shipped runtime behavior.

Co-authored-by: Codex <noreply@openai.com>