Pattern miner: Add backlog-drip archetype for the All You Can Eat scheduling pattern

[github-actions]

Caution

Protected files were modified in this change.
This pull request is in request_review mode and requires explicit human scrutiny before merge.

Protected files: .github/copilot-instructions.md

Mined upstream sources for recurring workflow shapes not yet captured in the pattern library.

Upstream evidence

• githubnext-agentics/files/workflows/issue-monster.md — schedule every 30m, skip-if-match on is:pr is:open is:draft author:app/copilot-swe-agent combined with skip-if-no-match: "is:issue is:open", serving one assignment at a time from a bounded backlog (issues labeled cookie).
• github-gh-aw/files/.github/aw/workflow-patterns.md — a dedicated, named section, "All You Can Eat Pattern" (lines 88-134), that documents the exact same shape as a recommended, reusable pattern: a scheduled workflow capped to one unconsumed open item (max: 1 + skip-if-match), with an explicit "Shape" YAML example, rules for stable-identity matching and starvation prevention via expires, and a "Learn from the closed output" section describing how the agent should read the previous item's close reason/comments to avoid re-proposing rejected work.

Two independent upstream demonstrations (one workflow example plus a full documented, named pattern) satisfy the "at least two upstream workflows, or one workflow plus explicit upstream documentation describing it as a recommended pattern" bar from the mining process.

Why the existing library did not already cover it

The current library already has skip-if-match/skip-if-no-match as tracked trigger-combo terms and generic dedupe tips scattered across several archetypes (daily-test-improver, code-improvement, performance-nut, etc.), but none of those describe the specific self-throttling "one unconsumed item at a time, learn from how the last one was closed" shape — a schedule that reactivates only once its own last output is consumed, with feedback-driven backlog curation via cache-memory/repo-memory. This is distinct from community-digest (periodic report tied to a fixed time window, not gated on consumption of the previous output) and from the generic pre-activation "too many open PRs" checks used by several other curated archetypes.

What changed in patterns/

• Added patterns/archetypes/backlog-drip.json — a new curated archetype (success_rate: null, count: 0) describing the "All You Can Eat" shape: schedule trigger, create-issue safe output/tool, tips covering max: 1 + skip-if-match on a stable identity marker, expires to prevent starvation, and reading closed-item history from cache-memory/repo-memory before proposing the next item.
• Appended "backlog-drip" to archetypes in patterns/manifest.json.
• Added the backlog-drip runtime rendering entry to patterns/workflow-generation.json (icon, capabilities, role-steps-style prompt body with process/constraints).
• Updated .github/copilot-instructions.md archetype counts (30 → 31 user-facing archetypes) and the curated-archetypes list to include backlog-drip, and updated the matching assertion in test/copilot-instructions.test.js (curated count 21 → 22) so the existing sync test continues to pass.

No other files were touched; metadata, anti_patterns, trigger_combos, research_findings, degraded_workflows, and top_repos were left untouched.

Validation

• npm test — 326 passed, 1 pre-existing failure unrelated to this change (describes the committed pattern-library corpus, not stale scan data fails on baseline main too, because the manifest's generated_at metadata has drifted past the date hard-coded in .github/copilot-instructions.md; reproduced by stashing this change and re-running the same test).
• npm run build — succeeds; dist/patterns/archetypes/backlog-drip.json is emitted and dist/patterns/manifest.json/dist/patterns/workflow-generation.json include the new entry.

Candidates deliberately left out

• Wiki-driven code sync (agentic-wiki-coder.md/agentic-wiki-writer.md, gollum trigger + repo-memory wiki backend): only one upstream workflow pair from a single repo demonstrates the gollum trigger, and no separate upstream doc recommends it as a standalone pattern (the wiki: checkout option and repo-memory wiki backend are documented as a memory-storage choice, not a workflow archetype) — did not meet the two-workflow/one-workflow-plus-doc bar.
• Mermaid-diagram-on-demand (archie.md): a single upstream workflow with no corroborating pattern documentation elsewhere.
• Assign-to-Copilot-agent backlog items (issue-monster.md, duplicate-code-detector.md, large-file-simplifier.md using assignees: copilot/assign-to-agent): already implicit in existing code-health-auditor tips ("Assign findings to copilot") and not a distinct enough archetype shape to warrant a new entry.
• hide-older-comments on add-comment: a small safe-output refinement seen across 5 upstream workflows, but it only affects an existing tip's wording on already-covered archetypes (performance-nut, daily-test-improver) and ranked below the "All You Can Eat" pattern for this run's single-archetype budget.

Generated by Pattern Miner · copilot · auto · 445.3 AIC · ⌖ 16.5 AIC · ⊞ 8.3K · ◷

[github-actions]

Protected files were modified in this pull request and require manual scrutiny before merge.

Please verify that each protected-file change is intentional, policy-compliant, and safe:

• Protected files: .github/copilot-instructions.md