Pattern miner: Add code-health-auditor archetype: bounded structural code-quality audits

[github-actions]

Upstream evidence

Mined githubnext-agentics/files/workflows/ (from data/import-sources.json's githubnext/agentics source):

• duplicate-code-detector.md — schedule/workflow_dispatch trigger, create-issue safe-output with expires: 2d, title-prefix, assignees: copilot, bounded to one recurring defect pattern per run.
• large-file-simplifier.md — schedule trigger with skip-if-match: 'is:issue is:open in:title "[large-file-simplifier]"', tracker-id: large-file-simplifier, create-issue with expires: 2d and assignees: copilot, bounded to the single largest file exceeding a size threshold.
• code-simplifier.md — schedule trigger with skip-if-match: 'is:pr is:open in:title "[code-simplifier]"', tracker-id: code-simplifier, create-pull-request with expires: 1d.

Confirmed as a deliberate, documented pattern (not a one-off) in github-gh-aw/files/.github/aw/syntax-agentic.md, which documents tracker-id as a first-class field ("tag all created assets ... enables searching and retrieving assets associated with this workflow") and in github-gh-aw/files/.github/aw/safe-outputs-content.md, which documents assignees: [user1, copilot] ("use 'copilot' for bot") as a supported create-issue/create-pull-request option.

Recurring pattern found

Three upstream workflows demonstrate the same shape: a scheduled, bounded structural code-health audit (oversized files, duplicated code, unnecessary complexity) that reports or fixes exactly one finding per run, guarded against re-triggering with tracker-id + skip-if-match (or deduplicate-style title checks) and expires on the created asset, with assignees: copilot so remediation can continue automatically.

Why the existing library did not cover it

• code-improvement is issue-driven, general-purpose bug/CI fixing — not scheduled structural auditing.
• linter-miner proposes new lint rules from recurring defects, not one-off structural findings (file size, duplication) reported directly as an issue/PR.
• performance-nut targets runtime/build performance, not structural code health (size, duplication).
• No existing archetype's tips mention tracker-id/skip-if-match dedup guards or assignees: copilot, both of which recur across all three source workflows and the upstream docs.

What changed

• Added patterns/archetypes/code-health-auditor.json — new curated archetype (success_rate: null, count: 0, top_repos: []) recommending schedule trigger, create-issue safe output, create-issue tool, phase-based prompt style, with tips on scoping to one structural dimension, using tracker-id/skip-if-match, expires, and assignees: copilot.
• Appended code-health-auditor to archetypes in patterns/manifest.json.
• Added the code-health-auditor runtime generation entry (icon, capabilities, instructions, prompt body) to patterns/workflow-generation.json.

No other files were touched.

Validation

• npm test — 183 tests passed across all 12 test files.
• npm run build — production build succeeded; new archetype JSON emitted correctly under dist/patterns/archetypes/.

Candidates deliberately left out

• assignees: copilot as a standalone tip on existing archetypes — folded into the new archetype's tips instead of touching unrelated curated entries, per the "keep changes minimal" rule.
• A separate "duplicate-detector" or "large-file" archetype — the three upstream workflows are different instances of one shape (bounded structural audit with dedup guards), so they were merged into a single code-health-auditor archetype rather than three near-duplicate entries.
• cost-tracker.md (workflow_run trigger) — only one upstream example, and workflow_run is already flagged as a documented anti-pattern (13% success) in the library; not enough corroboration to add as a new archetype.
• repository-quality-improver.md's cache-memory focus-area rotation — interesting but only one upstream example without corroborating documentation recommending it as a named pattern; left out per the two-workflow-minimum rule.

Generated by Pattern Miner · auto · 76.2 AIC · ⌖ 3.37 AIC · ⊞ 8.3K · ◷

[pelikhan]

@copilot these AWs should use LSP tool

[Copilot]

@copilot these AWs should use LSP tool

Implemented in 22f4c08 (with configuration validation in 6526fc1). Generated Copilot code-health auditor workflows now configure TypeScript/JavaScript LSP, use it before text search, and allow the Node registry needed to install the server.

[github-actions]

Reviewer: matt-pocock (TypeScript/JS-focused)

Reviewed the diff (new code-health-auditor archetype + lspFor() LSP wiring in src/js/workflow.js, plus the guard-fix commit validating command/fileExtensions before emitting LSP config).

No blocking issues found:

• lspFor correctly returns null for non-Copilot engines and now defensively filters out incomplete LSP entries (missing command or empty fileExtensions) before emitting YAML — good guard against malformed template data producing broken frontmatter.
• New archetype JSON and generation entry follow the existing schema/conventions used by sibling archetypes.
• Test suite passes locally (185/185, including the two new tests covering LSP frontmatter generation and the non-Copilot-engine no-op case).

Minor non-blocking observations (not requesting changes):

• lspFor's validity check only inspects the first LSP entry structurally; if patterns ever defines multiple languages per archetype with mixed validity, only the invalid ones are dropped, which is the intended behavior — confirmed correct, no action needed.
• The generated args array is serialized with simple string quoting ("${arg}" via template) rather than a YAML-safe serializer; fine today since values are static/curated JSON, but would be worth a sanitize/escape step if archetype data ever becomes user-editable.

No merge-blocking concerns.

Generated by Specialist PR Review for #157 · auto · 25.2 AIC · ⌖ 2.51 AIC · ⊞ 7.9K