Repository navigation
ClonePeek v0.5.0 — Reusable GitHub Action
ClonePeek v0.5.0 — reusable GitHub Action
ClonePeek is an offline, read-only execution-surface inventory for unfamiliar repositories. It examines modeled npm lifecycle scripts, VS Code folder-open tasks, devcontainers, Claude Code and Cursor hooks, MCP process configs, and bounded explicit local shell-script references. It does not execute repository code or prove that a repository is safe.
New: use ClonePeek directly in GitHub Actions
- uses: actions/checkout@v4
with:
fetch-depth: 0
- uses: harshitethic/clonepeek@v0.5.0
with:
path: .
fail-on: high
format: sarif
output: clonepeek.sarifThe composite action runs the scanner from the separate, versioned action checkout, using Python's standard library. It does not install dependencies from the repository being scanned. The workflow may fetch a Python runtime through GitHub's setup-python action; the scan itself makes no network requests.
Action capabilities
- Report-only by default (
fail-on: none); optional low, medium, high, or critical threshold. - Optional
changed-sinceGit ref and reviewed fingerprintbaseline. - Text, JSON, and SARIF reports; optional private output file instead of full console report.
- Machine-readable finding counts and a GitHub job summary with counts only.
- Runnable on Windows, macOS, and Linux.
Verification
- 64 passing Python unit tests.
- Separate reusable-action GitHub CI smoke tests pass on Ubuntu, macOS, and Windows.
- 6/6 unit-test CI jobs pass across operating systems and Python 3.10/3.13.
- Wheel and source distribution builds verified.
- The 51-fixture synthetic evaluation is not a representative estimate of security detection accuracy. Known false positives and misses remain documented.
Try the CLI directly
uvx --from git+https://github.com/harshitethic/clonepeek@v0.5.0 clonepeek /path/to/repoSee the GitHub Action guide, limitations and threat model, and community discussions. Contributions with harmless, redacted reproducible examples are welcome.
Important: A finding is a review prompt, not a malware verdict. Reports may include sensitive source excerpts; review them before making artifacts public.