Skip to content

docs(rfc): widen the crewmate prune rule to package-installed agents - #14908

Closed
buluoray wants to merge 1 commit into
mainfrom
docs/crewmates-prune-rule
Closed

buluoray wants to merge 1 commit into
mainfrom
docs/crewmates-prune-rule

Conversation

@buluoray

@buluoray buluoray commented Sep 29, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: the crewmates launch RFC records the startup prune rule that removes every never-chatted crewmate the old agent sync generated, including those bound to package-installed agents.

Screen 03 of rfc-crewmates-launch.md limits the one-time prune to rows bound to "a custom agent of the person's own — never ... a package's". The sync that generated those rows enrolled package-installed agents the same way, so on an install whose agents mostly come from packages the prune removes almost nothing and the Crewmates roster stays full of templates. The RFC also never says what "chatted with" means, which left the implementation counting any session that ran the agent.

Why it matters

#14897 implements the wider rule and is blocked by the First Principles lane until an accepted RFC on main records it.

Not a goal

What changed (motivation → approach → change)

Section 03 of docs/request-for-change/rfc-crewmates-launch.md now names the rule #14897 ships.

Before After
Candidate rows bound to the person's own agent bound to an installed spec of the person's own or of a package; never one of the runtime's own agents (a Kiro Crew-owned spec or a kirocrew stamp), a private copy, or an uninstalled spec
"Chatted with" undefined a turn in the crewmate's own Crewmates-page thread, live or archived
Created crewmates not stated a row with a member_id is never removed
Unreadable history nothing removed, pass re-runs that crewmate is kept and named in the marker; the pass finishes; a new marker name gives old installs one more run

The reasoning paragraph says why package rows count (the same sync made them, and one click re-enrols them), why only the crewmate's own thread counts, and why one bad file keeps only its own crewmate. Sections 6, 7 and 8 and the implementation-prs front matter match.

Backwards compatibility

Compatible: documentation only; no code, config or API changes.

Tests

N/A — documentation only; scripts/docs-lint.sh passes.

Manual verification

N/A — documentation only.

Related Issues

no linked issue: records the product decision #14897 implements.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@buluoray
buluoray requested a review from a team as a code owner September 29, 2026 03:01
@buluoray
buluoray requested a review from kyleseaman September 29, 2026 03:01
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 29, 2026
@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 19975b9c944f198842e1ba219d89d3f9e9b72121 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 19975b9

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 19975b9c944f198842e1ba219d89d3f9e9b72121: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — ✅ PASS

Design-level review of 19975b9c944f198842e1ba219d89d3f9e9b72121 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Confirm the 2026-09-28 product-owner revision is a real human decision — this PR reverses the RFC's earlier explicit "never a package's" exclusion.

[DESIGN-REVIEWED] 19975b9

Description read

[DESCRIPTION-READ] 3926f103b6d45313fb29e8f889f888938322b6f20daeadad023efdbb9c2cdf0a

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

Copy link
Copy Markdown
Contributor

Opus 5 Review — ✅ no blocking findings

Reviewed 19975b9c944f198842e1ba219d89d3f9e9b72121 — this comment is updated in place on each push.

Review details

The revised §03 excludes the runtime's agents by a source stamp none of their enrolled rows carry, so the exclusion is vacuous and contradicts §7 in the same diff.

FINDING — docs/request-for-change/rfc-crewmates-launch.md:161 — never the runtime's own \kirocrew` stampnames a discovery source that onlykirocrew.json, kirocrew-lite.jsonandkirocrew-guest.json carry (agent_discovery._global_agent_info), and _do_agents_sync skips exactly those (disc.source != "kirocrew"), so no such row can exist; the other eight OWNED_KIRO_AGENT_FILESspecs (worker, knowledge, research, heartbeat, the four conductors) declarename == stem, get source = "builtin"withkirocrew_owned=True, and WERE enrolled — the shipped pass excludes them by the second test spec.kirocrew_owned (crewmate_prune_migration.py:320-325), which the stamp wording drops, so an implementation of §03 as written prunes those eight roster rows while line 352 of this same diff asserts runtime-owned and spec-less rows ... are untouched → Fix: say "never a spec the runtime owns (kirocrew_owned— a filename inOWNED_KIRO_AGENT_FILES)" instead of naming the kirocrew` source stamp.

FINDING — docs/request-for-change/rfc-crewmates-launch.md:177 — the revised rule writes a new marker name, so an install the first pass left untouched gets one more run understates the effect: _prune_locked gates only on marker.exists() for the single PRUNE_MARKER path (crewmate_prune_migration.py:182,904-907), so a rename re-runs the pass on every install that already completed one, re-judging under the narrowed "chatted with" the rows pass 1 deliberately kept on session-metadata or activity-log evidence — which §7's on its first start and once says cannot happen → Fix: state that the new marker name re-runs the pass on every install and say what becomes of rows the first pass kept, or drop the "and once" claim in §7.

[OPUS-REVIEWED] 19975b9

Verdict parsed from the review's SHA-scoped output markers for commit 19975b9c944f198842e1ba219d89d3f9e9b72121.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 19975b9c944f198842e1ba219d89d3f9e9b72121: <one-sentence reason>

The launch RFC's screen 03 limited the startup prune to rows bound to
the person's own agents. The old sync enrolled package-installed agents
the same way, so those rows are the same leftover state and stay on the
roster forever under that rule.

Section 03 now covers package-installed specs, leaves the runtime's own
and uninstalled specs alone, defines "chatted with" as a turn in the
crewmate's own Crewmates-page thread, states that a created crewmate is
never removed, and keeps an unreadable thread's crewmate without holding
the rest of the pass back.
@buluoray
buluoray force-pushed the docs/crewmates-prune-rule branch from 19975b9 to d2b6faf Compare September 29, 2026 03:37
@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 29, 2026
@buluoray

Copy link
Copy Markdown
Contributor Author

Folded into #14897: the RFC §03 amendment now ships in the same commit as the prune change it records.

@buluoray buluoray closed this Sep 29, 2026
@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Sep 29, 2026
@bolichen97
bolichen97 deleted the docs/crewmates-prune-rule branch November 6, 2026 07:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant