Skip to content

[auto] reflect: narrow-the-scope names the shared-premise step when one gate rejects twice - #289

Open
EdbertChan wants to merge 1 commit into
stack/EdbertChan/reflect/pstack-premise-20260908/auto-reflect-corpus-scan-skips-slow-oversized--6b763dadfrom
stack/EdbertChan/reflect/pstack-premise-20260908/auto-reflect-narrow-scope-names-shared-premise--d0706777
Open

[auto] reflect: narrow-the-scope names the shared-premise step when one gate rejects twice#289
EdbertChan wants to merge 1 commit into
stack/EdbertChan/reflect/pstack-premise-20260908/auto-reflect-corpus-scan-skips-slow-oversized--6b763dadfrom
stack/EdbertChan/reflect/pstack-premise-20260908/auto-reflect-narrow-scope-names-shared-premise--d0706777

Conversation

@EdbertChan

@EdbertChanEdbertChan commented Sep 8, 2026

Copy link
Copy Markdown
Owner

Summary

The narrow-the-scope skill gains one method sentence: when one gate rejects a second edit, write down the premise both edits shared, then read the gate's source before a third.

The skill's fires_example fixture gains a matching scenario, which the skill test coverage gate requires for a prose-only skill change.

Review Claim

The new sentence names the missing step that ended two real three-attempt loops, in the file's existing tone, without changing any trigger.

Review Lane

behavior

Review Unit

product-skill

Safety Invariant

One sentence added to an existing numbered step and one fixture paragraph; no hook, install path, or trigger condition changes, so nothing fires differently.

Assumptions: headless mode, Safety Invariant is unconfirmed by a human.

Slice Rationale

Evidence: in two read-confirmed Invoker sessions the agent only escaped a three-attempt loop by reading the gate's own source (a body-keyword checker that scanned the whole PR body; a UNIT_PATTERNS regex), and the shared premise was never written down.

This is the prose item from the same reflect pass as the stack's first PR (corpus_scan fix). It is a different Review Unit, so it lands as its own PR on top.

Non-goals

  • No change to narrow-the-scope's mechanical trigger or engine/hooks/narrow-the-scope/.
  • No change to the stays_silent fixture or the reflect skill.

Test Plan

Test Plan
  • python3 scripts/check_skill_test_coverage.py --base origin/main --head HEAD -> ok skill test coverage
  • python3 engine/skills/make-pr/scripts/preflight.py --base origin/main -> ok preflight passed (codify-has-code, no-dated-provenance, three-harness, ecosystem-boundaries, skill-file-refs, skill-test-coverage, skill-trigger-mechanism all ok)
  • python3 scripts/check_no_dated_provenance.py --base origin/main -> ok no dated provenance
  • bash scripts/run_all_tests.sh -> exit 0 (31 suites OK)

Revert Plan

Revert Plan
  • Safe to revert? Yes
  • Revert command: git revert a30b78a
  • Post-revert steps: None
  • Data migration? No

Depends-On: #288

🤖 Generated with Claude Code

https://claude.ai/code/session_014NSMiMsTPuon8otdhRaJEq

…ne gate rejects twice
Add one sentence to the method: when the same gate rejects a second edited
version, write down the one-sentence premise both edits shared, then read
the gate's source before a third edit. In two read-confirmed Invoker
sessions the agent only escaped a three-attempt loop by reading the gate's
own source (a body-keyword checker that scanned the whole PR body; a
UNIT_PATTERNS regex), and the shared premise was never written down.
The fires_example fixture gains a matching scenario so the prose-only
skill's coverage tracks the new trigger.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014NSMiMsTPuon8otdhRaJEq
Change-Id: Id0706777811cc493b05bb45ec8a42b8f1a81fb9c
@EdbertChan

Copy link
Copy Markdown
OwnerAuthor

This pull request is part of a Mergify stack:

#Pull RequestLink
1[auto] reflect: corpus_scan skips slow or oversized transcripts instead of crashing#288
2[auto] reflect: narrow-the-scope names the shared-premise step when one gate rejects twice#289👈

@cursor

cursorBot commented Sep 8, 2026

Copy link
Copy Markdown

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_e8a3e704-efeb-4d59-8341-257bd808c366)

Sign up for freeto 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

@EdbertChan