clean-code-setup

Pass

Audited by Gen Agent Trust Hub on Sep 29, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends adding an external skill from a specific third-party GitHub user (dmmulroy/anti-slop) and suggests various linting and testing packages from public registries (NPM, PyPI, Cargo) that are not part of a pre-defined trusted vendor list.\n- [COMMAND_EXECUTION]: The hook scripts (commit-gate.sh, check-on-stop.sh) execute shell commands using bash -c. The commands are derived from project configuration files like package.json, Makefile, or justfile, which could be manipulated to execute arbitrary commands if the repository is untrusted.\n- [INDIRECT_PROMPT_INJECTION]: The installed hooks (format-and-lint.sh, check-on-stop.sh, commit-gate.sh) ingest session metadata and tool inputs from stdin. A malicious project configuration or crafted tool input could potentially influence the behavior of these gates.\n
  • Ingestion points: Reads project configuration files (package.json, pyproject.toml, Makefile, etc.) and parses session data (session ID, file paths, tool commands) from stdin in the hook scripts.\n
  • Boundary markers: None identified in the script logic for handling external configuration or session data.\n
  • Capability inventory: The hooks utilize bash, npx, git, ruff, mypy, go, and cargo to perform linting and checks.\n
  • Sanitization: The scripts perform basic parsing of JSON input using jq (if available) or sed for command extraction.\n- [DYNAMIC_EXECUTION]: The skill generates shell scripts in the .claude/hooks/ directory based on provided templates and configures the agent to execute them automatically during specific lifecycle events.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 29, 2026, 05:58 AM