The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python CLI scripts (scripts/version_bump.py and scripts/release_plan_scaffold.py) that execute the local git binary to retrieve commit history.
The scripts use subprocess.run with a list of arguments and shell=False (default), which prevents standard shell injection.
An automated scanner detection for 'remote code execution' was identified as a false positive; it was triggered by the word 'Fetch' in function docstrings referring to local data retrieval.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (commit messages, changelogs) which can contain untrusted content.
Ingestion points: Commit subjects/bodies via git log (scripts/version_bump.py, scripts/release_plan_scaffold.py), user-supplied CHANGELOG.md files (scripts/changelog_check.py), and JSON event data (scripts/dora_metrics.py).
Boundary markers: Absent; the scripts parse the raw text to extract semantic versions and change types.
Capability inventory: The agent can perform file writes, execute git commands, and summarize changelog data based on the ingested content.
Sanitization: The scripts use regex to validate specific formats (Conventional Commits, SemVer) but do not perform security-specific sanitization of the natural language content.
[SAFE]: The skill follows security best practices for secret management, specifically instructing users not to bake secrets into artifacts and to use secrets managers or environment variables.
No hardcoded credentials or obfuscated content were detected.
All logic is restricted to the Python standard library, requiring no third-party package installations.

release-engineering