Docs: architecture.md and the release-smoke recorder still understate the Participant roster

[dichovsky]

Two roster restatements that #26 did not name, found while implementing it (#56).

• docs/design/architecture.md:458-459 reads: "The start command is not a literal /crew everywhere. Claude and Gemini use /crew, Codex uses $crew, and in Copilot you run /agent, pick crew, and then type the prompt." Four participants that also return /crew are missing — src/platforms/antigravity.ts:29, pi.ts:51, little-coder.ts:40, opencode.ts:33. This restates the same fact as FR-G07, which docs: align every roster introduction on the eight Participant CLIs #56 corrected to enumerate all eight (docs/design/srs.md:513-515), so architecture (authority rank 4) is now out of step with the SRS (rank 2). The understatement is older than docs: align every roster introduction on the eight Participant CLIs #56 — architecture.md already omitted Pi, Little Coder, and opencode before it — and was left in place because Participant roster drift: README, CLAUDE.md/AGENTS.md, SRS, product-spec, and npm metadata disagree with the eight-CLI registry #26 did not cite this file and .github/ISSUE_TRIAGE.md's scope rule routes unnamed files to a new issue.

• tests/tools/release-smoke.test.ts:15 instructs the maintainer to "Run it (with the five Participant CLIs / backends installed)". The recorder iterates PARTICIPANT_TARGETS + BACKEND_TARGETS (line 24) — eight Participant CLIs and two Model Backends.

Impact: the architecture document contradicts a higher-authority requirement on the exact invocation fact FR-G07 governs, which is the disagreement docs/README.md's authority order exists to prevent; a reader taking architecture.md at face value would conclude that six of the eight participants have no documented start command. The release-smoke instruction understates what a maintainer must install before a release smoke run, so the run can be recorded as complete while three participants and both backends were never exercised.

Acceptance: architecture.md:458-459 enumerates the same eight invocations as FR-G07 and stays consistent with it ("Claude, Gemini, Antigravity, Pi, Little Coder and opencode use /crew, Codex uses $crew, and in Copilot you run /agent, pick crew, and then type the prompt"); release-smoke.test.ts:15 says eight Participant CLIs and two Model Backends. No code change is required for either — both are prose, and the registry is already correct.

[dichovsky]

Triage: re-verified against main @ d311767. Confirmed with a correction — both passages reproduce at the cited lines; one supporting claim in the body is premature.

• docs/design/architecture.md:458-459 reads exactly as filed: "Claude and Gemini use /crew, Codex uses $crew, and in Copilot you run /agent, pick crew, and then type the prompt." Checked against the registry, six platforms return /crew, not two — claude.ts, gemini.ts:34-36, antigravity.ts:28-30, pi.ts:50-52, little-coder.ts:39-41, opencode.ts:32-34. codex.ts:29-31 returns $crew; copilot.ts:33-35 returns the /agent selection form. So four of the eight invocations are missing from the sentence.
• tests/tools/release-smoke.test.ts:15 reads "Run it (with the five Participant CLIs / backends installed)". The recorder iterates PARTICIPANT_TARGETS and BACKEND_TARGETS (:24 import, used in the body) — eight Participant CLIs and two Model Backends, ten targets.

Correction to the body. It says architecture.md "is now out of step with the SRS (rank 2)" because #56 corrected FR-G07 to enumerate all eight. That has not happened yet: PR #56 is open and unmerged, and FR-G07 on current main (docs/design/srs.md:510-512) still lists seven, omitting Antigravity. Today architecture.md is out of step with the registry, and FR-G07 is out of step with it too — but only by one platform, where architecture.md is out by four. The rank-2-versus-rank-4 contradiction the Impact paragraph describes becomes true when #56 merges; it is not the present state. This does not change the fix or its acceptance.

Labels applied:

• documentation — the architecture.md half is prose in docs/, with the code and registry already correct.
• bug — the release-smoke.test.ts:15 half is not in docs/; it is a source comment contradicting the code directly beneath it, which the triage table routes to bug. Same class as Console: sidebar docstrings say five view links while it renders six #53 and Console: server.ts header comment omits the FR-U36 agent-archive POSTs from its "EXACTLY" route list #60.
• good first issue — two single-line prose edits with the correct text already spelled out in the Acceptance and independently confirmed above against the registry. No design judgment.

Not a duplicate. #26 (PR #56 open) cites README, CLAUDE.md/AGENTS.md, the SRS, product-spec, and npm metadata, and neither of these two files; #42 is docs-site prose; #35 is architecture.md §4.1's source tree, a different section from §8's setup narrative. Different stale facts in a shared file are not duplicates of each other.

---

Generated by Claude Code