Pattern miner: Add rejection-learning and cross-repo pattern tips from upstream mining

[github-actions]

What was mined

Mined githubnext/agentics (61 workflow files under workflows/) and github/gh-aw (docs under .github/aw/) for recurring shapes not yet captured by patterns/.

Recurring pattern found

"All You Can Eat" sequential-backlog gating with rejection learning. A scheduled workflow keeps at most one open output alive (skip-if-match/tracker-id with max: 1 and expires), and before proposing new work it reads the outcome of its most recently closed item (close reason, review comments) to avoid re-proposing rejected ideas.

Evidence — 3 upstream workflows demonstrate this shape:

• githubnext-agentics/files/workflows/code-simplifier.md (skip-if-match: 'is:pr is:open in:title "[code-simplifier]"', create-pull-request with expires: 1d)
• githubnext-agentics/files/workflows/large-file-simplifier.md (tracker-id: large-file-simplifier, skip-if-match, create-issue with expires: 2d)
• githubnext-agentics/files/workflows/issue-monster.md (skip-if-match/skip-if-no-match gate, learns from prior assignment outcomes)

Corroborating documentation — github-gh-aw/files/.github/aw/workflow-patterns.md, section "All You Can Eat Pattern" (lines 88-133), explicitly names this as a recommended design with a "Learn from the closed output" sub-section describing exactly the rejection-learning step (search recently closed items, read close reason, persist accept/reject history in cache-memory).

Additionally found a smaller recurring shape: cross-repository monitoring where a workflow reads from a target-repo/TARGET_REPOSITORY variable and writes labels/comments back to that external repo via target-repo on safe outputs (githubnext-agentics/files/workflows/contribution-check.md), corroborated by github-gh-aw/files/.github/aw/workflow-patterns.md section "Cross-Repository Pattern" (lines 284-293).

Why the library did not already cover this

• code-improvement and code-health-auditor already had cache-memory rotation tips (avoiding repeating the same category/dimension) but nothing about learning from how a previously closed item was received — a different, complementary signal.
• No archetype mentioned target-repo or cross-repo write configuration; contribution-guidelines-checker only covered same-repo PR checks.

What changed in patterns/

Small, additive tip refinements only (no new archetype, no triggers/safe-outputs/success_rate/count/top_repos/metadata changes):

• patterns/archetypes/code-improvement.json: added a tip about capping to one open PR (skip-if-match/max: 1/expires) and learning from recently closed PRs' rejection signals before proposing a new fix.
• patterns/archetypes/code-health-auditor.json: added a tip about reading recently closed findings' close reason/review feedback before creating a new one.
• patterns/archetypes/contribution-guidelines-checker.json: added a tip about monitoring an external repository via target-repo on safe outputs, GitHub toolsets, and token-scope documentation.

Validation

• npm test: 327 tests passed (19 test files).
• npm run build: succeeded, produced dist/ including all patterns/archetypes/*.json.

Candidates deliberately left out

• Weekly repo tree map / weekly issue activity charts (weekly-repo-map.md, weekly-issue-activity.md, repo-chronicle.md) — all three use upload-asset for charts/visuals, but this is already represented by the existing extras.charts entry in workflow-generation.json and the community-digest archetype's create-discussion guidance; not a new gap.
• AI Moderator's skip-roles/skip-bots/lock-for-agent combo (ai-moderator.md) — only one upstream workflow demonstrates it and I could not find corroborating upstream documentation naming it as a distinct recommended pattern (only generic field docs in syntax-core.md), so it does not meet the two-workflow-or-documented-pattern bar.
• assign-to-agent auto-triage (issue-monster.md, duplicate-code-detector.md, large-file-simplifier.md) — while assign-to-agent appears 3 times, it's already listed as a safe output in the existing issue-triage configuration profile in manifest.json, so it isn't a net-new capability gap.
• New standalone archetype for repo-status/team-status/repo-chronicle "daily activity narrative issue" — this shape is already fully covered by the existing status-report and community-digest archetypes (issue vs. discussion output), so a new archetype would be a near-duplicate.

Generated by Pattern Miner · copilot · auto · 259 AIC · ⌖ 10.8 AIC · ⊞ 8.7K · ◷

[github-actions]

Reviewer: ponytail (content/pattern-data change — additive JSON tips only, no code/UI/TS changes to review from a Matt Pocock angle)

Reviewed the three tip additions in patterns/archetypes/*.json:

• code-health-auditor.json: rejection-learning tip (read close reason/review feedback before creating a new finding)
• code-improvement.json: caps to one open PR + rejection-learning tip
• contribution-guidelines-checker.json: cross-repo monitoring via target-repo tip

All three edits are valid JSON, additive only, and consistent in tone/structure with existing entries in the same arrays. No triggers, safe-outputs, success_rate, or count fields were touched, matching the PR description. The guidance is plausible and well-scoped (each tip names a concrete mechanism: skip-if-match/max/expires, target-repo, closed-item lookback).

No blocking issues found — the changes are self-contained documentation/pattern-data additions with no runtime or UI impact.

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• github.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "github.com"

See Network Configuration for more information.

Generated by Specialist PR Review for #278 · copilot · auto · 19.1 AIC · ⌖ 5.3 AIC · ⊞ 7.2K