Capture PR #969 UMS lessons

[d-morrison]

Closes #1020

Summary

• Add a gh pr list --state merged gotcha to memories/github.md.
• Record that a routine main merge can create cross-file duplication in a prose PR.
• Point redundancy review at the existing normalized overlap instrument instead of literal grep.
• Record that a reviewer's corrected citation is itself a factual provenance claim.

Existing coverage checked

• memories/github.md had related mergedAt guidance, but not the gh pr list --state merged ordering trap.
• shared/workflow/sync-with-main.md had adjacent merge-collision guidance, but not clean merges creating a DRY violation across files.
• scripts/find-near-duplicates.py already provides normalized shingle-style overlap detection, so the new entry says to use the existing instrument.
• shared/workflow/address-every-comment.md covered wrong suggestions, but not corrected citation provenance.

Verification

• python3 scripts/validate-skills.py -> valid, with the pre-existing uninitialized sembr-skills submodule warning.
• python3 scripts/check-memory-file-size.py -> no memory file exceeds 1200 lines.
• NLB_BASE_REF=origin/main python3 /Users/ezramorrison/Documents/GitHub/gha/check-new-line-breaks/check-new-line-breaks.py -> no lines missing semantic breaks.
• Banned-punctuation scan over git diff -U0 origin/main...HEAD added lines -> examined 113 added lines, 0 hits.

[Copilot]

Pull request overview

Records reusable lessons from PR #969’s review lifecycle.

Changes:

• Documents merge-created prose duplication and normalized overlap checks.
• Adds citation-provenance verification guidance.
• Records merged-PR query ordering pitfalls.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
shared/workflow/sync-with-main.md	Covers duplication introduced by clean merges.
shared/workflow/challenge-redundant-content.md	Adds normalized overlap-check guidance.
shared/workflow/address-every-comment.md	Requires verification of replacement citations.
memories/github.md	Documents merged-PR listing pitfalls.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[claude]

Code review

Reviewed at head 3743b4b1000095c527dd84fd6752562ad5583127. This PR adds four documentation-only entries to memories/github.md, shared/workflow/address-every-comment.md, shared/workflow/challenge-redundant-content.md, and shared/workflow/sync-with-main.md, capturing UMS lessons from PR #969's review lifecycle (issue #1020).

Verification performed:

• Confirmed the gh pr list --state merged ordering claim directly: gh pr list --repo Morrison-Lab/ai-config --state merged --limit 8 --json number,mergedAt returns PRs in descending-number order with non-monotonic mergedAt (fix(gip): spell the base branch as <default-branch> throughout #1013 at 05:36:51Z appears before ums: a background watcher reports failure as silence by default #1012 at 05:45:06Z), exactly matching the case record.
• Cross-checked every cited PR number (workflow: batch merge and resolve, always #969, workflow: question the assignment, not only the claims #971, ums: seven verification learnings, each a check that certified nothing #966, fix: preview-by-default and diff-scoped writes for semantic-line-breaks.py, plus a rule against asserting corpus gaps from a grep #951, ums: short CI failures are not a credential signature, and a handed premise is still a claim #955, ums: dispatch on "I owe you", and monitor claims by type rather than by confidence #947, ums: capture PR 966 lifecycle lessons #1019, fix(gip): spell the base branch as <default-branch> throughout #1013, ums: a background watcher reports failure as silence by default #1012) and issue Capture PR #969 ARDI lessons in ai-config #1020 against live repo state — all exist with the states/dates/file-touch sets the prose claims (verified via gh pr diff --name-only for each).
• Ran scripts/find-near-duplicates.py against shared/workflow/*.md at thresholds down to 0.08 — no new duplicate pairs introduced (the PR practices what it preaches).
• Ran scripts/validate-skills.py and scripts/check-memory-file-size.py — both pass, matching the PR's own verification section.
• Scanned added lines for banned punctuation (em/en-dash, curly quotes, multiplication sign) — none found.
• Manually checked added lines for the one-clause-per-line / one-sentence-per-line convention — compliant.
• All four additions follow the corpus's Do/Don't pairing convention (CLAUDE.md's "Record both the pattern and the anti-pattern") with concrete, verified case records.

One finding posted inline (memories/github.md:95-97): the recommended replacement command gh search prs --merged=">=<date>" doesn't work — --merged is a boolean flag ("Filter based on merged state"), not a date filter. The correct flag is --merged-at date. Confirmed by running both forms against the live repo. This is the one place in the diff where a claim wasn't independently verifiable by inspection alone and turned out to be wrong — everything else checked out.

Verdict

Needs minor changes — one real bug (a non-functional CLI flag in the new memory entry's own recommended fix) to correct; everything else in the diff is accurate, well-cited, and compliant with the repo's own conventions.

[claude]

Code review

Reviewed at head 1909bd5f3aa985bc69b14b8c7bcd6d122508ec4e (previous review was at 3743b4b1). Since that review, commit e7ba1517 addressed both outstanding findings, plus a main merge with no conflicts.

Prior finding resolved: the previous review and a Copilot review both flagged that memories/github.md's recommended replacement command used --merged (a boolean flag) instead of --merged-at (the date-range flag). This is now fixed at both occurrences (lines 95 and 97), and I re-verified the corrected form actually works:

$ gh search prs --repo Morrison-Lab/ai-config --merged-at ">=2026-08-01" --limit 5 --json number,state
[{"number":1019,"state":"merged"}, ...]

Copilot's second finding also resolved: shared/workflow/challenge-redundant-content.md's new section originally implied scripts/find-near-duplicates.py's default run could certify a narrow single-file audit clean. The fixed text now correctly frames the script's default corpus (skills/*/SKILL.md) and default threshold/top-N (0.05 / 25) as tuned for broad candidate-listing, not as a clean-result certificate for a narrow finding -- verified against the script's own DEFAULT_CORPUS, DEFAULT_THRESHOLD, and --top values.

Re-verification performed:

• Confirmed both flag-syntax fixes work against the live repo (above).
• Confirmed scripts/find-near-duplicates.py's actual defaults match what the corrected prose now claims about it.
• Re-ran python3 scripts/validate-skills.py and python3 scripts/check-memory-file-size.py at the current head -- both pass.
• Re-ran python3 scripts/find-near-duplicates.py --corpus "shared/workflow/*.md" -- 0 pairs above threshold across 34 units, no new duplication introduced.
• Scanned all added lines (full PR diff) for banned punctuation (em/en-dash, curly quotes, multiplication sign) -- none found.
• Re-checked the address-every-comment.md and shared/workflow/sync-with-main.md additions, which the fix commit did not touch -- unchanged from the previously-verified content.
• Confirmed issue Capture PR #969 ARDI lessons in ai-config #1020 (which this PR closes) matches the PR's stated scope.

No new findings.

Verdict: Ready for merge.

[github-actions]

Claude finished review — View run

---

Posted successfully.

Summary

I reviewed PR #1021 at head 1909bd5f3aa985bc69b14b8c7bcd6d122508ec4e, which is one commit ahead of the previously-reviewed head (3743b4b1). The intervening commit (e7ba1517) fixed both outstanding findings:

1. My prior blocking finding — memories/github.md's recommended gh search prs --merged=">=<date>" command (a non-functional boolean flag) — is now corrected to --merged-at ">=<date>" at both occurrences. I re-ran the corrected command against the live repo and confirmed it works.
2. Copilot's finding on shared/workflow/challenge-redundant-content.md (about find-near-duplicates.py's default run being oversold as a clean-result certificate) is also fixed, and I verified the corrected prose accurately reflects the script's actual defaults (DEFAULT_CORPUS, DEFAULT_THRESHOLD=0.05, --top=25).

I re-ran validate-skills.py, check-memory-file-size.py, and find-near-duplicates.py against the current head (all pass, no new duplication), scanned the full diff for banned punctuation (none found), and confirmed the untouched files (address-every-comment.md, sync-with-main.md) were already verified accurate in the prior round.

Verdict

Ready for merge.

[github-actions]

💰 Cost: $6.8083 (review) — run

[d-morrison]

Ready for merge.

Validation at current head 1909bd5f:

• python3 scripts/validate-skills.py passed; existing sembr-skills uninitialized submodule warning only.
• python3 scripts/check-memory-file-size.py passed.
• NLB_BASE_REF=origin/main python3 /Users/ezramorrison/Documents/GitHub/gha/check-new-line-breaks/check-new-line-breaks.py passed.
• Banned-punctuation scan examined 120 added lines and found 0 hits.
• Targeted markdownlint passed.
• GitHub checks are green, latest claude review at 1909bd5f says Ready for merge, and all review threads are resolved.