[bruno-archastro]

Summary

• Teach capability-gated archdev log knowledge and artifact-backed archdev log finding in the canonical skill and monitor reference.
• Cover bounded lexical recall, omitted history/content, attribution and local verification; preview then deliberate --publish; evidence/applicability; and linked updates or corrections without erasing history.
• Preserve the top-level-only sharing rule, optional no-finding outcomes, authorization/secrets checks, and the distinction between session-selected projects and a shared checkout.
• Check advertised commands and flags independently. Older CLIs retain reviewed log search / log post fallbacks without forced upgrades, hook changes, or bypassing a supported command's rejected lookup.
• Regenerate both documents in all eight plugin bundles and add two generator regression tests. Correct blanket delivery/project-validation statements that do not apply to the new read-only or preview paths.

Verification

• npm run typecheck
• npm test: 127 passed, including both new documentation/distribution regressions.
• npm run check: 148 generated files match.
• Canonical skill/reference match all eight generated copies byte-for-byte; skill descriptions remain within 1,024 characters.
• git diff --check
• Read-only help checks: the installed 0.54.6 CLI does not advertise these commands; the companion implementation build exposes the documented commands and flags.

Rollout and scope

• Companion implementation: https://github.com/ArchAstro/firstlanding/pull/19406. Guidance is safe to ship before that CLI release because each new capability is optional and independently checked.
• No installed skills/hooks, plugin wiring, catalog/package versions, or unrelated risk/rules guidance changed.
• Documentation/tests do not establish live agent uptake or verified reuse. Real-harness evaluation remains follow-up work; conversation delivery into coding sessions is separate.

Agent session: 01a15ec9-6f14-7246-b49e-6fc94f9f2984

[bruno-archastro]

Reviewed the published assessments for the exact PR head c894e8c76294592d478e254d59fe725e482fa7b7: PR risk is Medium; the six available code-region assessments are Low. The concerns are about CLI-contract evidence and mirrored guidance, not a demonstrated contradiction.

Verification behind the guidance:

• The two added generator regressions check advertised-command/flag gating, safe fallback boundaries, preview versus explicit publication, required evidence/project fields, public message IDs, omission/uncertainty language, and exact distribution to all eight harnesses.
• All 127 generator tests pass; all 20 GitHub CI checks pass on this head. The generator's existing byte-for-byte distribution and description-length tests also pass.
• Read-only help checks confirm the installed 0.54.6 CLI lacks finding/knowledge, while the local companion implementation advertises all documented flags. Its command tests check that previews perform reads only, publication uses the normal lesson/delivery path, rejected projects/linked messages do not publish, and recalled posts remain untrusted information.

The companion implementation is now available for independent review in firstlanding#19406, head 18fbbddfda0be3940415e2acce7b27596a9c16b5: command definitions and preview/publication checks, command and session-routing regressions, and hook integration. Its refreshed focused verification has 222 passing tests. The author approved a one-commit local Aster bypass for the documented baseline-only sandbox failures; normal hosted CI remains the gate for that implementation. These checks do not establish live agent uptake or verified reuse; that remains follow-up work. No change was made solely to lower the server grade.