Pattern miner: Mine upstream patterns: chart digests, agent-delegation, and PR throttling tips

[github-actions]

Mined upstream githubnext/agentics workflow samples and github/gh-aw reusable-guidance docs for recurring shapes not yet reflected in the curated pattern library, then added three small, evidence-backed tip refinements.

Sources mined

• /tmp/gh-aw/data/upstream/githubnext-agentics/files/workflows/*.md (58 workflow files)
• /tmp/gh-aw/data/upstream/github-gh-aw/files/.github/aw/*.md (68 reusable-guidance docs)

Recurring patterns found and why they weren't already covered

1. Chart-and-discussion digests (community-digest)

• Evidence: repo-chronicle.md and weekly-issue-activity.md both combine upload-asset with create-discussion to publish Python-generated trend charts embedded via returned URLs; charts.md and charts-trending.md document this as the recommended pattern for chart-bearing periodic reports.
• Gap: the community-digest archetype already recommended create-discussion for narrative digests but had no tip on how to add charts to that narrative — charts were only exposed as a generic "extra" capability, disconnected from the digest use case. Two upstream workflows plus two supporting docs corroborate the shape (exceeds the 2-workflow threshold).
• Change: added one tip to patterns/archetypes/community-digest.json describing the upload-asset + pre-step chart-generation + embed-by-URL flow.

2. Scheduled backlog item handed to an implementation agent (backlog-drip)

• Evidence: issue-monster.md pairs a scheduled, one-item-at-a-time trigger shape (already the shape captured by backlog-drip) with the assign-to-agent safe output (allowed: [copilot], ignore-if-error: true) instead of merely creating an issue for a human. safe-outputs-automation.md and designer-mappings.md document assign-to-agent as the general-purpose way to route work to the Copilot coding agent, corroborating this as a supported, recommended pairing rather than a one-off.
• Gap: backlog-drip only recommended create-issue; nothing in the library described the alternative of auto-delegating the served item to an agent.
• Change: added one tip to patterns/archetypes/backlog-drip.json describing the assign-to-agent addition and its ignore-if-error safeguard.

3. Pre-activation PR-count throttle for proactive maintainer workflows (repo-maintainer)

• Evidence: repo-assist.md, ci-coach.md, efficiency-improver.md, and glossary-maintainer.md all gate scheduled runs behind a steps:-level check that counts open PRs sharing the workflow's title prefix and skips the run past a threshold (MAX_OPEN_PRS), to avoid spamming maintainers. repo-maintainer's own anti_patterns list already cites repo-assist three times as an observed failure mode, and other archetypes in the library (code-improvement, daily-test-improver, documentation-updater, link-checker, linter-applier, performance-nut) already carry this exact tip — repo-maintainer was the one proactive, PR-creating, schedule-capable archetype missing it.
• Change: added the same standard tip (worded identically to its use elsewhere in the library) to patterns/archetypes/repo-maintainer.json.

What changed in patterns/

• patterns/archetypes/community-digest.json: +1 tip (chart embedding via upload-asset)
• patterns/archetypes/backlog-drip.json: +1 tip (assign-to-agent delegation)
• patterns/archetypes/repo-maintainer.json: +1 tip (pre-activation PR-count throttle)

No new archetype was added, manifest.json/workflow-generation.json were left untouched (the refinements only edit tips fields), and metadata, anti_patterns, trigger_combos, research_findings, degraded_workflows, success_rate, count, and top_repos were not modified anywhere.

Validation

• npm test: 325 passed, 2 failed — the 2 failures (test/copilot-instructions.test.js) are pre-existing on main before this change (verified via git stash), caused by a stale generated_at date / archetype count mismatch unrelated to this diff.
• npm run build: succeeded (✓ built in 154ms), all three edited archetype JSON files copied into dist/patterns/archetypes/ with updated sizes.

Candidates deliberately left out

• Multi-device/responsive docs testing (multi-device-docs-tester.md): only one upstream workflow demonstrates this shape with no corroborating doc recommending it as a named pattern — below the two-workflow threshold for a new archetype.
• CI cost/efficiency optimization (ci-coach.md, efficiency-improver.md): two upstream workflows exist but they target different KPIs (CI runtime/cost vs. energy/computational footprint) and neither is documented as a named pattern in patterns.md's pattern index; the existing code-improvement and agent-cost-tracker archetypes already cover adjacent ground, so a new archetype risked redundancy without stronger evidence.
• Discussion-mining into issues (discussion-task-miner.md): only one upstream workflow; no second example or explicit doc recommendation found.
• Mermaid diagram generation on demand (archie.md): only one clear standalone example (lean-squad.md/agentic-wiki-writer.md/tech-content-editorial-board.md use mermaid incidentally within much larger, already-distinct archetypes, not as the primary shape) — did not meet the corroboration bar for a dedicated archetype or tip.
• label_command as a primary trigger for an archetype: documented in workflow-patterns.md and triggers.md, but no upstream agentics workflow uses it as its sole/primary trigger, so there isn't a concrete workflow shape to attach a refinement to yet.

Generated by Pattern Miner · copilot · auto · 362.8 AIC · ⌖ 9.82 AIC · ⊞ 8.3K · ◷