Optimization: This skill uses on-demand loading. Heavy content lives in references/ and is loaded only when needed.

Language rule: Skill instructions are written in English. When generating user-facing output (reports, files, documentation), detect the project's primary language (from README, comments, docs, commit messages) and produce the output in that language. If the project is bilingual, ask the user which language to use before proceeding.

CI Pipeline Optimizer — Biomimetic

"The leafcutter ant colony has no project manager. Yet it optimally relocates 15% of all tropical vegetation."

The 9 biological models → CI patterns

Read references/patterns.md for detailed explanations, biology, and dual GitLab/GitHub examples.

Workflow for analyzing an existing pipeline

Step 1 — Map the real DAG

Draw the actual dependencies. Identify sequential jobs that could run in parallel. Compute the critical path.

Step 2 — Leafcutter audit

Does each job do ONE thing? Are artifacts properly defined (stigmergy)? Are there generalist jobs that should be split?

Step 3 — Slime mold audit (Physarum)

Are there jobs running without any relevant change? Are cache keys based on content? Are retries selective? Content-hashed keys and reproducibility are shared in ../shared/determinism.md. CI stages map onto the shared T0-T4 rungs (../shared/gate-ladder.md): the release gate fails at the lowest red rung.

Step 4 — Army ants audit

Which slow jobs can be fanned out in parallel? Are runners ephemeral? Are fan-ins happening too early?

Step 5 — Honeybee audit

Are runners sized for the right profile? Is scale-to-zero enabled? Do critical jobs have dedicated runners?

Step 6 — Mycelium audit

Are there SPOFs (registry, cache, single runner)? Does the cache have a waterfall fallback? Do multi-project pipelines share expensive artifacts?

Step 7 — Immune system audit

Is there fuzzing on parsers, serialization, auth? Do tests use property-based testing? Is the crash corpus persisted across runs?

Step 8 — Spore audit

Does the matrix cover OS × arch × version × features? Are nightly/beta combinations marked allow_failure? Are impossible combinations excluded?

Step 9 — Tardigrade audit

Are there tests under degraded network conditions? Tests under resource pressure? Tests with clock skew? Is graceful degradation verified?

Mandatory rules

GitHub Action version verification

Before recommending a version bump for a GitHub Action (e.g. @v3 → @v4), verify that the target tag actually exists:

1. Floating major tags (@v4) are a convention, not a requirement. Some maintainers don't create them.
2. Always verify via gh api repos/{owner}/{repo}/git/refs/tags/{tag} that the exact tag exists.
3. If the floating major tag doesn't exist, pin to the latest patch version (e.g. @v4.1.1).
4. Never assume a major tag exists just because patch tags do.

Reference incident: bumping cosign-installer@v3 → @v4 broke CI because the v4 tag didn't exist (2026-03-27).

Gotchas

Read ../gotchas.md before producing output to avoid known mistakes.

Anti-patterns to identify

Success metrics

• Critical path: < 50% of the theoretical sequential total time
• Cache hit rate: > 80% on dependencies
• Runner utilization: > 70%
• Flaky test rate: < 1%
• SPOF count: 0

Pipeline Scoring (15 dimensions)

Read references/scoring.md for detailed scoring criteria and scorecard template.

Target score: > 45/60 (75%) for a production project.

Pipeline Pre-Mortem

Before merging a pipeline change, imagine the failures:

1. Poisoned cache (malicious artifact)
2. Compromised runner (secret leak)
3. Registry down (no fallback)
4. Flaky job retry-masked for 3 months
5. Cache miss → 45-minute pipeline
6. Cross-workflow dependency silently broken
7. DAG dependency diamond (C fails silently)

For each scenario: what mitigation ALREADY exists? If none → accepted risk, document it.

Pipeline Mutation Testing

The mutation set + scorecard threshold is this skill's post-verification gate; the 3-phase definition-of-done structure (pre → during → post) is canonical in ../shared/done-gate.md.

If > 2 mutations pass silently, the pipeline has holes.

Reference templates

• references/patterns.md — The 9 biological models in detail with GitLab + GitHub Actions examples
• references/scoring.md — Detailed 15-dimension scoring + scorecard
• references/gitlab-ci-biomimic.yml — Complete GitLab pipeline
• references/github-actions-biomimic.yml — Complete GitHub Actions pipeline
• references/runner-autoscaler.toml — GitLab Runner autoscaler config (honeybees)
• references/cache-strategy.md — Advanced cache strategies (slime mold + mycelium)

Dynamic Handoffs

Rule: Recommend, don't auto-execute.

Integration with other cli-* skills