Index of every page on this site: llms.txt
All skills
Run it yourself

Audits a repository's GitHub Actions workflows for CI optimization opportunities — missing caches, redundant setup, sleep-based readiness, long test jobs without sharding, full-history checkout, dead env vars, build-cache misconfig, and ~60 more patterns across caching, redundancy, parallelization, conditional execution, trigger scope, and hidden failures. Use when: (1) analyzing a repo's CI for optimization opportunities, (2) producing a prioritized report with measured wall-clock and runner-minute savings, (3) re-auditing after upstream CI changes. Do not trigger for: general CI setup help, writing new workflows from scratch, non-GitHub-Actions CI systems, or security/posture audits — use `ci-secure` for those.

Use this Skill: https://skilld.dev/gh/starslingdev/skills/ci-speedup

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencespatternsINDEX.md

≈3.4k tokens on demand. Your agent reads this file only when SKILL.md points to it.

ci-speedup pattern index

One line per pattern: OPT<n> — title (impact, class) — what it is about. Find the pattern here, then read opt<n>-*.md in this directory in full. Never open ../optimization-patterns.md: it is the generated public copy of these files. This order is also the compiled file's order; contributors see README.md.

Category 1 — Caching

  • OPT1 — Unnecessary Tool Install (MEDIUM, static) — e.g. Playwright browsers where unused
  • OPT2 — Uncached Large Downloads (MEDIUM, static) — browser or toolchain downloads uncached
  • OPT3 — Read-Only Turbo Cache (never populated) (MEDIUM, static) — TURBO_CACHE remote:ro never written
  • OPT4 — Docker Layer Cache Missing (MEDIUM, static) — docker build without cache-from
  • OPT5 — pnpm Store Not Cached (or Wrong Setup Order) (MEDIUM, static) — pnpm store uncached or setup order
  • OPT6 — Cache Key Entropy Too High or Unstable (MEDIUM, static) — run_id, sha or date in cache key
  • OPT7 — pnpm Version Drift Across Workflows (LOW, static) — different pnpm versions per workflow
  • OPT8 — Cache Key Granularity Mismatch (MEDIUM, static) — coarse hashFiles key over app caches
  • OPT9 — Tool-Specific Cache Flag Not Enabled (HIGH, static) — ESLint, Jest, tsc cache flags off
  • OPT79 — A Cache That Costs More Than It Saves (MEDIUM, data-driven) — cache restore slower than the install
  • OPT93 — NuGet Restores to a Slower Disk on Windows Runners (MEDIUM, data-driven) — measured cache placement on Windows
  • OPT96 — Cache Never Saved on the Default Branch (MEDIUM, static) — PR caches lack a reachable default-branch save
  • OPT101 — Cache and Build Use Different Directories (MEDIUM, data-driven) — missing cache path on every successful save attempt
  • OPT112 — Restored Go Build Cache Trimmed Before Reuse (HIGH, data-driven) — the same exact Go cache archive makes the job measurably slower once it is past cmd/go's five-day trim limit

Category 2 — Redundancy

  • OPT11 — Redundant Environment Variables (LOW, static) — same env var at two levels
  • OPT12 — Duplicated Setup Across Jobs (MEDIUM, static) — copy-pasted setup steps
  • OPT13 — Build Step in Jobs That Don't Need It (MEDIUM, static) — lint/typecheck job runs a build
  • OPT14 — Repeated Checkout/Setup Without Artifact Handoff (MEDIUM, static) — every job rebuilds; no artifact handoff
  • OPT15 — Cross-Workflow Build Redundancy (HIGH, static) — same build in several PR workflows
  • OPT16 — Within-Job Duplicate Commands (LOW, static) — same command twice in one job
  • OPT110 — Uploaded Artifact Has No Repository Consumer (MEDIUM, static) — upload name is never downloaded or consumed by release/deploy
  • OPT77 — Repeated Fixed Setup Across Independent Small Jobs (MEDIUM, data-driven) — small checks each pay full setup
  • OPT105 — Local Hook Work Is Repeated in Pull-Request CI (MEDIUM, data-driven) — measured PR steps repeat a resolved local-hook command
  • OPT113 — A Test Fixture Is Much Larger Than Its Contract Needs (MEDIUM, static) — test archives or hashes a full executable when bytes suffice

Category 3 — Docker

  • OPT17 — Sleep-Based Container Readiness (MEDIUM, static) — sleep instead of healthchecks
  • OPT18 — All Containers Started for Single-Service Tests (MEDIUM, static) — compose up starts every service
  • OPT19 — Test Source Sleep Dominance (includes Playwright waitForTimeout) (HIGH, static) — fixed sleeps / waitForTimeout in tests
  • OPT20 — Unpinned Docker Image Tags (LOW, static) — docker images on :latest
  • OPT102 — Foreign-Architecture Image Compilation Runs Under Emulation (HIGH, static) — multi-platform Dockerfile compiles without a native or cross target
  • OPT114 — Docker Image Layers Repeat npm Advisory Lookups Already Run by CI (MEDIUM, static) — host audit retained; image-layer duplicates disabled

Category 4 — Parallelization

  • OPT21 — Unnecessary needs: Dependencies (MEDIUM, static) — needs: on a job it does not use
  • OPT22 — Sequential Workflows via workflow_run (MEDIUM, static) — workflow_run chains instead of jobs
  • OPT23 — Single-Threaded Matrix (max-parallel: 1) (MEDIUM, static) — matrix max-parallel: 1
  • OPT24 — Long Test Job Without Sharding (HIGH, static) — long test job, no shards
  • OPT25 — Shard Imbalance (MEDIUM, data-driven) — one shard far slower than the rest
  • OPT88 — Test Concurrency Is a Guess, Not a Measurement (MEDIUM, static) — unmeasured test worker count
  • OPT89 — The Slowest Platform Compiles Its Own Tests (HIGH, static) — cross-compile once, test on each platform

Category 5 — Actions and Checkout

  • OPT26 — Outdated Action Major Versions (LOW, static) — old actions/* major versions
  • OPT27 — Duplicate setup-node Calls in Same Job (LOW, static) — setup-node twice in a job
  • OPT28 — Full Git History Checkout (MEDIUM, static) — fetch-depth: 0 where not needed
  • OPT76 — Submodule / Git LFS Checkout Payload (MEDIUM, static) — unused submodules or LFS objects
  • OPT80 — Checkout Stalls on the Tail (MEDIUM, data-driven) — git fetch stalls on slow runs

Category 6 — Conditional Execution

  • OPT29 — Merge Queue Skip at Step Level Only (MEDIUM, static) — merge_group skip at step level
  • OPT30 — Matrix Jobs Without Job-Level Conditional (MEDIUM, static) — matrix runners provisioned for nothing
  • OPT31 — Conditional Step With Unconditional Setup (MEDIUM, static) — setup runs when its consumer is skipped
  • OPT109 — Failure Reporter Runs on Healthy Pull Requests (MEDIUM, static) — always-running report-only job allocates a runner on green runs

Category 7 — Trigger and Scope

  • OPT32 — Missing paths/paths-ignore on Expensive Workflows (HIGH, static) — no path filters on expensive workflows
  • OPT33 — No Draft PR Gating on Expensive Jobs (MEDIUM, static) — expensive jobs run on draft PRs
  • OPT34 — No Changed-Package Filtering (MEDIUM, static) — monorepo runs every package
  • OPT35 — Missing fail-fast on Non-Diagnostic Matrix Dimensions (LOW, static) — fail-fast: false on shard matrices
  • OPT36 — Cron Schedule Too Frequent (LOW, static) — cron runs too often
  • OPT37 — Workflow Trigger Dependency Gap (Cache Race) (HIGH, static) — cache writer and reader race
  • OPT38 — Non-Content Trigger Event Types (PR edited, etc.) (MEDIUM, static) — PR edited/labeled events re-run CI
  • OPT39 — Multi-Language Matrix Without Path Filter (MEDIUM, static) — scanner matrix without path filters
  • OPT40 — Monorepo Job Runs Regardless of Affected App (MEDIUM, static) — app job runs when another app changed
  • OPT108 — Comment Filter Runs After the Runner Starts (HIGH, static) — comment body or author is filtered only after runner allocation
  • OPT111 — Automated Code Review Runs on Every Pull Request Push (MEDIUM, static) — fix-up pushes repeat a full automated review
  • OPT90 — Every Pull Request Pays for Coverage That Could Run After Merge (HIGH, data-driven) — full matrix on PRs, subset would do
  • OPT107 — Default-Branch Full CI Runs Too Often to Stay Current (HIGH, data-driven) — dense post-merge validation can batch against the current tip

Category 8 — Release Workflow

  • OPT41 — TURBO_FORCE: true Disabling All Caching (HIGH, static) — TURBO_FORCE in release
  • OPT42 — TURBO_CACHE: remote:rw in Release (HIGH, static) — TURBO_CACHE remote:rw in release

Category 9 — Queue Times and Concurrency

  • OPT43 — Excessive Queue Time (MEDIUM, data-driven) — jobs wait long for a runner
  • OPT97 — Queue Caused by a One-at-a-Time Resource (MEDIUM, data-driven) — queue grows with in-flight jobs while execution stays flat
  • OPT44 — Concurrency Group Too Restrictive (MEDIUM, static) — concurrency group too narrow
  • OPT45 — Missing Concurrency Groups (HIGH, static) — no concurrency group
  • OPT46 — Superseded Runs Not Cancelled (Missing Concurrency or cancel-in-progress: false) (MEDIUM, data-driven) — superseded runs keep running
  • OPT47 — Redundant push + pull_request Double-Trigger (MEDIUM, data-driven) — push and pull_request both fire
  • OPT48 — High Job-Level Failure Rate (>15% over 30d) (MEDIUM, data-driven) — job fails over 15% of runs
  • OPT98 — The Merge Queue Re-proves What Is Already Proven (MEDIUM, data-driven) — strict-rule churn and repeated full merge-queue validation

Category 10 — Timing Anomalies

  • OPT100 — Infrastructure Faults Fail the Whole Run (MEDIUM, data-driven) — classify first-step infrastructure failures before retrying
  • OPT49 — Slow Setup Step (MEDIUM, data-driven) — cut; id retired
  • OPT50 — Post Steps Taking Too Long (MEDIUM, data-driven) — cut; id retired
  • OPT51 — Install-to-Test Ratio >50% (MEDIUM, data-driven) — cut; id retired
  • OPT81 — The Same Job Is Measurably Faster on Another Runner (MEDIUM, data-driven) — job faster on another runner class

Category 11 — Stack-Specific

  • OPT52 — Turbo Tasks Missing outputs in turbo.json (MEDIUM, static) — turbo task without outputs
  • OPT53 — Unstable Env Vars Invalidating Turbo Cache (MEDIUM, static) — changing env vars in turbo hash
  • OPT54 — Full-Repo pnpm -r Where Package Filters Are Possible (MEDIUM, static) — pnpm -r where a filter would do
  • OPT55 — vitest Running in Watch/Dev Mode in CI (MEDIUM, static) — vitest watch mode in CI
  • OPT56 — Playwright Traces/Videos Uploaded Unconditionally (MEDIUM, static) — traces/videos on passing tests
  • OPT57 — Missing timeout-minutes on Known-Flaky Integration Jobs (MEDIUM, data-driven) — no timeout on flaky integration jobs
  • OPT58 — Turbo Tasks Missing inputs in turbo.json (MEDIUM, static) — turbo task without inputs
  • OPT59 — Runtime-Only Env Vars in Turbo globalEnv (MEDIUM, static) — runtime-only vars in globalEnv
  • OPT60 — Turbo CI Configuration Missing (LOW, static) — turbo CI settings missing
  • OPT82 — Lint Builds the Whole Type Graph (MEDIUM, data-driven) — type-aware ESLint on a slow lint job
  • OPT103 — Lighthouse CI Audits Every URL N Times Per Pass, One After Another (MEDIUM, data-driven) — passes x URLs x runs, audited one after another
  • OPT106 — Runtime TypeScript Transpilation on the Long Pole (HIGH, data-driven) — a heavy long-pole step runs TypeScript through a runtime loader

Category 12 — Build Caching (language-agnostic)

  • OPT61 — Missing Dependency Caching (HIGH, static) — compiled-language deps uncached
  • OPT62 — Build Artifacts Destroyed Before Every Run (HIGH, static) — rm -rf of build dirs every run
  • OPT63 — Dependency Install with Cache Disabled (MEDIUM, static) — --no-cache installs on persistent runners
  • OPT86 — CI Repeatedly Compiles Dependencies from Source (HIGH, data-driven) — measured vendored builds or confirmed prebuilt-package opportunities
  • OPT91 — The Compile Is the Long Pole and Nobody Profiled It (HIGH, data-driven) — slow compile, profile it first; static Xcode flavor flags an unread index store or a generic simulator destination
  • OPT94 — The .NET SDK Can Reconcile Unrelated Machine-Wide Workloads (MEDIUM, data-driven) — benchmark a pinned isolated SDK
  • OPT104 — Rust Coverage Loads Unoptimized Debug Artifacts (MEDIUM, structural) — measured coverage pole uses Cargo's debug/unoptimized test profile

Category 13 — Hidden Failures and Dead Config

  • OPT64 — Repeated Workflow Attempts From Same Failing Job (LOW, data-driven) — re-runs of the same failing job
  • OPT65 — Billing Rounding Waste from Tiny Matrix Legs (LOW, data-driven) — many sub-minute matrix legs
  • OPT66 — SKU Arbitrage Ceiling from Expensive Hosted Runners (LOW, data-driven) — removed; id retired
  • OPT68 — Broken Step Masked by continue-on-error (MEDIUM, static) — continue-on-error hides a broken step
  • OPT69 — Dead Workflow Env Vars / Config (LOW, static) — env vars or config nothing reads

Category 14 — Structural / Critical-Path Levers

  • OPT70 — Scope the Build/Test to Only What Changed (HIGH, structural) — test only what changed
  • OPT71 — Expensive Non-Required Check on the Critical Path (HIGH, structural) — non-required check on the critical path
  • OPT72 — Redundant-Work Ratio (build/install ≫ payload) (MEDIUM, structural) — build/install dwarfs the tests
  • OPT73 — Shared Sub-Step Across Critical-Path Jobs (cluster-floor lever) (HIGH, structural) — same step re-runs across parallel jobs
  • OPT74 — Trust-Boundary-Forced Cold Work (producer/consumer split) (MEDIUM, structural) — fork PRs cannot warm the cache
  • OPT75 — Long Pole: Optimize or Relocate the Dominant Step (HIGH, structural) — one step dominates the long pole
  • OPT78 — Per-File Test Isolation Rebuilding Shared Module State (HIGH, structural) — per-file isolation re-imports the app
  • OPT83 — Independent Steps on the Long Pole Run One After Another (MEDIUM, structural) — independent steps run serially
  • OPT84 — A Slow Step on the Long Pole Runs a Tool That Has a Native Drop-in (MEDIUM, structural) — Prettier/ESLint/tsc with native drop-ins; npm/pnpm/Yarn install → bun install
  • OPT85 — Repeated Work Runs in Many Processes Instead of Bounded Batches (HIGH, structural) — repeated process or worker-pool startup per item
  • OPT87 — Tests Wait on a Slow Dependency They Do Not Verify (HIGH, structural) — tests call a slow real dependency

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 alert2d3 checks · Risk SAFE
  • Gen Agent Trust Hub2d

    The skill is a robust CI optimization tool that audits GitHub Actions workflows. It identifies performance bottlenecks using deterministic logic and provides analysis of job logs. While it processes untrusted log data, it includes strong defensive instructions to mitigate indirect prompt injection. A minor best-practice issue regarding unsafe YAML loading was identified in a utility script.

  • Socket2d

    No alerts

  • Snyk2d

    Risk: CRITICAL · 2 issues

Signed by skilld at 9a84115. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 16 hours ago.

Activeupdated yesterday

README badge

README badge for starslingdev/skills/ci-speedup