Ci local
Run the checks a GitHub Actions workflow would run, locally, when Actions is unavailable or out of quota. Parses .github/workflows/*.yml, extracts the jobs/steps that gate merges (lint, typecheck, test, build), translates them to local commands respecting the workflow's pinned node/python versions and env, executes them sequentially, and reports a parity table of what passed locally vs. what can't be replicated (service containers, secrets, matrix dimensions) and why. Activates on "CI quota", "Actions is down/unavailable", "run CI locally", "verify like CI would", "run the pipeline on my machine", or a pre-push request to check a branch the way CI checks it. Not for generating a new workflow file (use ci-generate), not for debugging why a specific CI run failed on GitHub (use fix-bug or review-code), and not for the release/hotfix gating process itself (use gitflow), this skill only produces the local stand-in when the real thing isn't reachable.From its SKILL.md
npx -y skills add mgiovani/cc-arsenal --skill ci-localAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 8 stars8 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
6.8 KB, as published. Nobody here has run it
CI Local
Reproduce a GitHub Actions run on the local machine. The workflow YAML is the spec: read it, don't guess at what "the lint step" or "the test step" means.
Scope
GitHub Actions only (.github/workflows/*.yml). If the repo uses GitLab CI, CircleCI,
or Jenkins instead, say so and stop: no translation layer exists for those yet
(add one if it comes up twice, not speculatively).
Phase 1: Find the merge-gating jobs
- If invoked with
[workflow-file]and/or--job name, skip the glob: read only that file, and if--jobis given, extract only that job. OtherwiseGlobfor.github/workflows/*.yml(and.yaml). If none exist, tell the user there's no workflow to mirror and stop. - Read each file. A job gates merges if its workflow triggers on
pull_requestorpushto a protected branch: ignore jobs that only run onschedule,workflow_dispatch, orreleaseunless the user asks for those specifically. - For a multi-file or multi-job repo, use a haiku
Exploresubagent to read all workflow files in parallel and return a structured list of: job name, trigger, runs-on, steps (withuses/run/env/working-directory),services:,strategy.matrix, and any${{ secrets.* }}references. Keep this out of the main thread: workflow YAML is verbose and you only need the extracted summary. No subagent tool available? Read the workflow files directly and extract the same summary inline.
Phase 2: Translate steps to local commands
Walk the steps in order and translate each one. Do not invent a generic
lint && test && build: use exactly what the workflow does.
| Workflow step | Local translation |
|---|---|
actions/checkout | no-op: you're already in the working tree |
actions/setup-node with node-version: X | fnm use X (or nvm use X) before the next steps; if neither is installed, warn and continue on whatever node -v reports |
actions/setup-python with python-version: X | uv python pin X / uv run --python X ...; same fallback-and-warn if uv isn't available |
run: <cmd> with a working-directory: or env: block | run <cmd> from that directory with those env vars exported for just that command |
cache steps (actions/cache) | no-op: local disk cache already exists |
a run: step gated by if: on OS or event | skip if the condition can't hold locally (e.g. runs-on: windows-latest step on a Mac), and say so |
If no explicit version is pinned in the workflow, check .nvmrc / package.json#engines.node
or .python-version / pyproject.toml#requires-python before falling back to whatever's
on PATH.
Can't be replicated, flag, don't fake:
services:(Postgres, Redis, etc.): note the service and image; only attempt it if Docker is available and the user wants the extra step, otherwise mark the steps that depend on it as skipped.${{ secrets.* }}: check if a local.envsupplies the same variable name; if not, mark the step as skipped with the missing secret name, never substitute a fake value.strategy.matrix: run the one combination that matches the local machine (current node/python/OS); list the other matrix entries as not covered.
Phase 3: Execute sequentially
Run the translated commands via Bash, in the same order the workflow declares them,
stopping to report clearly the moment one fails (don't silently keep going past a
failed lint step and call the run "done"). Capture stdout/stderr for the report.
Phase 4: Report the parity table
CI Local Parity — <workflow file>, job "<job name>"
Step Local result Notes
---------------------------------------------------------------
checkout n/a already in working tree
setup-node 20 ok fnm use 20
lint (eslint) PASS
typecheck (tsc) PASS
test (vitest) FAIL 2 tests failing, see output above
build SKIPPED not run, blocked by failing test
postgres service NOT REPLICABLE no service container locally (Docker not requested)
deploy (secrets.AWS_*) NOT REPLICABLE secret not present in .env
Any version-fallback warning from Phase 2 (pinned node/python version unavailable,
falling back to whatever's on PATH) must surface as a caveat/NOT REPLICABLE note in
this parity table: never absorbed into a PASS.
Every row's result comes from a command you actually ran this session: never write PASS, FAIL, or a test count you didn't observe in the Phase 3 output.
State plainly at the end whether the branch would pass the real CI gate, and what's still unverified because it couldn't run locally.
Examples
Targeted job, --job lint given: skip the glob, read only the named job's steps
from the file the user pointed at, translate and run just those, report a parity
table scoped to that one job.
Python repo, no pinned version in the workflow: actions/setup-python has no
python-version: key → check pyproject.toml#requires-python, find >=3.11, run
uv run --python 3.11 pytest. Note the fallback source in the parity table
("version from pyproject.toml, not workflow"), don't silently treat it as pinned.
Matrix build, strategy.matrix: node: [18, 20, 22]: run once on whatever fnm/nvm
resolves locally (say 20), mark 18 and 22 as NOT REPLICABLE: matrix entry not run locally in the table instead of guessing they'd also pass.
Notes
- This is read-only with respect to git: no commits, no pushes, no workflow file edits.
Installing dependencies (
npm ci,uv sync, etc.) as part of a step is expected and fine. - If the same repo asks for this repeatedly, that's a signal to fix the actual CI quota/outage, not to keep leaning on the local stand-in: mention that once, don't nag.
What ships with it: 2 files
5.0 KB alongside SKILL.md
evals/
- evals.json2.9 KB
- trigger-eval.json2.1 KB
Gives 0 of the 12 instructions most quality gates skills give
Counted across 1,524 of the 2,830 authors here whose files we hold, read 2026-09-06
- Read full output and check exit codein 45 of 1524, across 40 files
- Verify output confirms the claimin 44 of 1524, across 39 files
- Identify the command that proves the claimin 43 of 1524, across 39 files
- Execute the full verification commandin 36 of 1524, across 30 files
- Produce a verification reportin 34 of 1524, across 18 files
- Review git diff changesin 30 of 1524, across 16 files
- Fix build failures immediatelyin 29 of 1524, across 9 files
- Group findings by severityin 28 of 1524
- State claim only with evidencein 27 of 1524, across 22 files
- Verify regression tests with red-green cyclein 26 of 1524, across 22 files
- Run the full test suitein 26 of 1524, across 25 files
- Run test suite with coveragein 25 of 1524, across 10 files
Said here and by no other author read
- Read workflow YAML files as the source of truth
- Extract merge-gating jobs triggered by pull requests or pushes
- Translate workflow steps to local commands sequentially
- Use fnm or nvm for node version management
- Export env vars for specific command execution
- Skip steps with conditions that cannot hold locally
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.