[skill-optimizer] Daily Skill Optimizer Improvements - 2026-07-20

[github-actions]

Summary

• Run mode: dry-run
• Status: ⚠️ Execution skipped — OPENROUTER_API_KEY not set; no benchmark data collected

Key Findings

1. AGENTS.md skill routing covers only ~20% of available skills — 40 skills exist under .github/skills/ but the routing table in AGENTS.md explicitly names only ~8 intent mappings. Skills like awf-release-integrator, console-rendering, copilot-review, debugging-workflows, go-codemod, prompt-token-efficiency, temporary-id-safe-output, and 20+ others have no routing entry. Agents cannot discover them without already knowing the path.

   • Expected impact: Agents will reliably load the right skill on first invocation instead of guessing, reducing hallucinated commands and missed conventions.

2. Several skills lack routing entries or have no intent trigger in AGENTS.md — go-linters/SKILL.md, reporting/SKILL.md, jqschema/SKILL.md, and 25 others have no corresponding line in the AGENTS.md routing table. Without an intent-matched entry, the lazy-load policy fails silently and agents skip specialized guidance.

   • Expected impact: Skill discovery via Agent Finder and ARD search becomes reliable; agents following the lazy-load policy will find the correct skill without exhaustive scanning.

3. The skill optimizer workflow runs permanently in dry-run mode because OPENROUTER_API_KEY is unavailable, yielding no graded benchmark data. The suite-results/ directory is never populated, so quality regressions in skill content go undetected across daily runs.

   • Expected impact: Enabling benchmark mode would produce graded pass/fail metrics per skill, giving concrete evidence for future optimizer runs and closing the feedback loop.

Evidence from Artifact

summary.json

{
  "repository": "github/gh-aw",
  "run_mode": "dry-run",
  "run_status": 0,
  "run_url": "https://github.com/github/gh-aw/actions/runs/29715631435"
}

run.log

dry-run: Docker available but OPENROUTER_API_KEY not set; skipping suite execution

Skill routing gap — present in .github/skills/ but missing from AGENTS.md routing:
awf-release-integrator, console-rendering, copilot-review, custom-agents, debugging-workflows, error-messages, error-pattern-safety, gh-agent-session, gh-agent-task, github-copilot-agent-tips-and-tricks, go-codemod, go-linters, http-mcp-headers, javascript-refactoring, jqschema, messages, optimize-agentic-workflow, otel-queries, pr-finisher, pr-to-go-linter, prompt-token-efficiency, reporting, sergo-examples, skillz-integration, ssl, temporary-id-safe-output, workflow-step-summaries (27 of 40 skills unrouted)

Small skill files with no routing — confirmed via wc -l:

• jqschema/SKILL.md: 23 lines
• go-linters/SKILL.md: 28 lines
• reporting/SKILL.md: 33 lines

Recommendations

1. Expand the AGENTS.md routing table to cover all 40 skills. Add one-line intent mappings for each unrouted skill (e.g., Go linter creation/validation → .github/skills/go-linters/SKILL.md). Group related skills using the wildcard pattern already used for github-*-query. This is a single-file edit to AGENTS.md with immediate discoverability impact for all agents.

2. Audit and populate missing description frontmatter and routing entries for every skill. Run grep -rL "^description:" .github/skills/*/SKILL.md to identify gaps. Add a one-sentence description capturing the intent trigger phrase so ARD search can match it. Every skill file should have a non-empty description in its YAML frontmatter.

3. Add OPENROUTER_API_KEY to repository Actions secrets and switch the skill-optimizer workflow to run_mode: benchmark. Once the secret is available, update the daily skill-optimizer workflow to enable graded suite execution. This populates suite-results/ with concrete pass/fail quality metrics and enables the optimizer to detect regressions in skill content.

Generated by ⚡ Daily Skill Optimizer Improvements · 30 AIC · ⌖ 8.56 AIC · ⊞ 4.3K · ◷

• expires on Jul 26, 2026, 7:59 PM UTC-08:00