Release Prep

Deep code audit + documentation sync + release preparation for Python packages. Use when preparing a release, checking code quality before publishing, auditi...

This skill appears to perform exactly the kinds of checks you want for a Python release, but there are important omissions and risks to consider before installing/allowing it to run: - Tools and dependencies: SKILL.md calls python, pytest (and pytest-cov), ruff, mypy, and standard Unix utilities (grep/sed). The package metadata lists no required binaries — ensure these tools exist in the execution environment before running. - Automatic fixes and publishing: 'fix' and 'release' modes will modify repository files and may tag/publish packages. Run first in 'audit' mode only, review any suggested fixes, and require manual approval before letting it perform commits, tags, pushes, or PyPI uploads. - Credentials: publishing to PyPI and pushing tags typically requires a PyPI API token and git credentials. The skill does not declare nor request these; do not provide credentials implicitly. If you enable 'release' features, supply tokens via a secure mechanism and restrict their scope. - Autonomous invocation: because the agent can be invoked autonomously, avoid granting it unchecked permission to run 'release' mode. Prefer running the skill interactively or in a sandbox/CI environment where you can review changes and control credentials. If you want to proceed: (1) run in audit mode first, (2) verify tools are installed, (3) review all diffs before accepting fixes, and (4) only provide publish credentials in a controlled, scoped way (e.g., CI secrets, short-lived token).

Type: OpenClaw Skill Name: release-prep Version: 1.0.0 The skill is classified as suspicious due to multiple critical shell injection vulnerabilities in SKILL.md. User-controlled content from project files (e.g., Python function names, dependency names, pyproject.toml entry points) is used in unquoted shell commands (e.g., `grep -rn "$func" src/`, `grep -rn "import $pkg\|from $pkg" src/`, `grep -q "$ep" README.md`). This allows an attacker to achieve arbitrary command execution (RCE) on the agent's host by crafting malicious project files. Additionally, there are significant prompt injection risks against the AI agent, particularly in phases involving report generation, auto-fixing, and changelog generation, where manipulated input could lead the agent to perform unintended actions or publish compromised packages.