The Agent Skills Directory

[SAFE]: The skill is a local auditing tool that implements robust internal security controls to analyze potentially untrusted repositories safely.
[COMMAND_EXECUTION]: The tool executes system utilities (git, gh, pytest, npm, go, cargo, make, devcontainer) using a restricted process launcher defined in process.py. This launcher uses a Python shim to perform an fchdir into a confined directory and scrubs the environment variables to prevent injection. Execution is limited to a hardcoded allowlist of tools and commands.
[DATA_EXFILTRATION]: The tool includes a dedicated output finalization layer in model.py that automatically redacts sensitive data patterns, including PEM private keys, bearer tokens, GitHub tokens, AWS keys, and absolute filesystem paths. Network activity is confined to the official GitHub API via the gh CLI.
[DYNAMIC_EXECUTION]: The engine includes a 'T3' execution tier in exec.py for running repository tests. This capability is disabled by default, requires an explicit user opt-in flag, and operates within an isolated temporary worktree copy with a restricted environment.
[PRIVILEGE_ESCALATION]: The safe_io.py module explicitly prevents privilege escalation or workspace escape by resolving and pinning the physical workspace boundary and addressing all repository content through directory file descriptors with no-follow semantics.

ra1-report