Local-first repository audits

Check before
you commit.

Catch risky workflows, configuration drift, and leaked credentials from your terminal, CI pipeline, or AI assistant.

Node 20+ · MIT license · Two runtime dependencies
v0.1.0 · CLI, Composite GitHub Action & MCP Server
example audit / terminal
$ npx ci-guard-mcp scan .

.github/workflows/deploy.yml:8
ERROR workflow/script-injection
  Untrusted event data in a shell script.
  Pass it through a quoted env variable.

src/config.ts:12
WARN env/undocumented
  API_URL is absent from an env template.

→ 2 findings. No source uploaded.
Illustrative output · values redacted
01 / Read-only by design
02 / Deterministic local checks
03 / CLI + Action + MCP
Four focused audit engines

Small checks. Useful signals.

Concrete findings with file locations, rule IDs, and a suggested next step. No account or cloud scanner required.

01 / WORKFLOWS

Keep CI from becoming an entry point.

Spot privileged PR checkouts, mutable action references, missing token permissions, and event expressions interpolated into shell scripts.

workflow/pwn-request
02 / ENVIRONMENT

Make the required configuration visible.

Compare literal environment reads across JS/TS, Python, Go, and Rust against templates, Docker ENV, and Compose declarations.

env/undocumented
03 / SECRETS

Catch credentials before they travel.

Recognize common token formats, private keys, and contextual high-entropy assignments. Scan actual staged blobs and redact every detected value.

secret/high-entropy
04 / DEPENDENCIES

Keep installs predictable.

Find competing lockfiles within a package, unencrypted dependency URLs, and unbounded version selectors such as * and latest.

lockfile/conflict
Meet your existing workflow

One engine. Three entry points.

Run directly via npx in your terminal, wire the composite action into GitHub Actions CI, or connect over stdio MCP.

TERMINAL

Run a local check.

Audit any repository in text, JSON, SARIF 2.1.0, or GitHub annotations format.

npx ci-guard-mcp scan . \
  --fail-on warn
npx ci-guard-mcp scan . \
  --format sarif > findings.sarif
CLI usage ↗
GITHUB ACTION

Give each PR a guard.

Annotations, a readable job summary, and finding counts from a composite action.

- uses: carelessdept/ci-guard-mcp@release-0.1.0
  with:
    path: .
    fail-on: warn
Integration details ↗
AI ASSISTANTS

Audit before an agent commits.

Four read-only tools over stdio. Paths stay within the selected launch root.

npx -y ci-guard-mcp --mcp \
  --root /path/to/repo

# scan_repo
# check_secret_leaks
MCP configuration ↗
Clear scope. Reviewable code.

A guardrail,
with visible boundaries.

Static checks cannot prove a repository is safe. Dynamic environment reads, encoded credentials, and remote workflow bodies need other review. Test fixtures and templates are excluded from secret scanning by default.

Read the detection limits ↗
  • No repository scripts are executed during scanning.
  • Source code stays on the machine running the audit.
  • Findings never contain detected secret values.
  • Ignore policy and official-action allowlists are configurable.
  • Excess-size files produce explicit warnings.
  • Deterministic JSON and SARIF 2.1.0 output for CI pipelines.