feat(agent): add the feedback and bug-report recipes, and rename the invited survey's recipe to rule-feedback

[theCodeDrift]

Stack (root → tip):

• docs(openspec): propose feedback-channels #477
  • feat(feedback): key surveys by kind and discriminate the payload on it #478
    • feat(feedback): send each kind to its survey, fill a bug's version info, and point to issues when telemetry is off #479
      • ➡️ feat(agent): add the feedback and bug-report recipes, and rename the invited survey's recipe to rule-feedback #480 (you are here)
        • feat(skill): route feedback and bug reports to their recipes #481
          • docs(openspec): archive feedback-channels #482

• agent/feedback.md is renamed to agent/rule-feedback.md (topic v4), carrying main's review flow, and the invite points to agent rule-feedback.
• New agent feedback and agent bug-report recipes. Each tells the agent to keep secrets, paths and unshared code out of the payload, to show the user the full payload and send only on an explicit yes, and to relay the issues URL when telemetry is off. Each embeds only its own schema branch.
• taskless agent gains a "Feedback recipes" section listing feedback and bug-report. The feedback command itself and rule-feedback stay unlisted.

This stack lands with gh stack merge, which merges every PR to main in one all-or-nothing operation, so no layer reaches main on its own. That matters because the layers only work together. For example, #478 requires kind on the payload while the recipe the invite names still describes a single survey, and #480 lists bug-report in the index before the skill routes to it.

[theCodeDrift]

@claude /review

[github-actions]

Claude finished @theCodeDrift's task in 4m 8s —— View job

---

Review complete

Review mode: incremental — read 0 prior review thread(s) before reviewing (.prior-review.json had no reviews, review threads, or comments beyond the trigger comment itself — this is the first review of this PR).

• Read .prior-review.json (empty — first review)
• Fetched PR diff, metadata, changed files (gh pr diff, gh pr view)
• Read all changed files in full, plus related files the diff interacts with but doesn't show (commands/feedback.ts, schemas/feedback.ts, survey/constants.ts, survey/invite.ts, prompts/recipes.ts, prompts/index.ts, commands/agent.ts, openspec/changes/feedback-channels/design.md)
• Reviewed for correctness, security, performance, test adequacy, clarity
• Posted inline comments
• Posted this summary

Overall: The three-channel split (rule-feedback / feedback / bug-report) is implemented consistently across the schema (schemas/feedback.ts), the survey registry (survey/constants.ts), the command (commands/feedback.ts), the recipe→schema wiring (prompts/recipes.ts), and the agent index (commands/agent.ts). Cross-references between the three recipes and feedback-invite.md are all updated correctly (verified every agent feedback → agent rule-feedback replacement landed, no stragglers). SURVEYED_TOPICS correctly excludes feedback/bug-report so they never carry the invite. This matches the stated design in openspec/changes/feedback-channels/design.md, and the tasks.md checklist is fully checked off and matches what's implemented.

Two findings, both [New]:

1. packages/cli/src/agent/rule-feedback.md (telemetry-off note is stale) — carried over verbatim from the old feedback.md, it still says a disabled-telemetry send just "exits 0 with nothing sent," but feedback send now redirects every kind (including rule) to https://github.com/taskless/cli/issues when telemetry is off — confirmed by this change's own design.md ("for every kind") and by the actual SEND_DISABLED string in commands/feedback.ts. The sibling feedback.md/bug-report.md recipes added in this same PR correctly describe the redirect (and are tested for it); rule-feedback.md isn't.
2. packages/cli/test/feedback-recipes.test.ts:187-192 — the "carries no survey invite, even with the gate open" test doesn't actually test with the gate open: the suite runs with telemetry disabled, which short-circuits surveyGateIsOpen before the SURVEYED_TOPICS check the test claims to be verifying. It would pass for any topic regardless of survey membership, so it wouldn't catch a future regression that accidentally surveyed feedback/bug-report.

Nothing else stood out — the zod strictObject/discriminated-union design, the per-kind TOPIC_INPUT_SCHEMAS embedding (verified each recipe only gets its own branch's keys via the new describe.each test), and the bug survey's CLI-filled version-information handling (bugVersionInformation, no network call, no identity-revealing fields) all look correct.

[theCodeDrift]

Re: @github-actions — "Claude finished @theCodeDrift's task in 4m 8s"…
#480 (comment)

Both findings are addressed in b48d57b; see the replies on their threads.

— AI Coding Agent