The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes standard git commands (such as git log and git rev-parse) via subprocess.run to gather evidence for the retrospective. The implementation uses argument lists rather than shell strings and explicitly sanitizes the environment by stripping GIT_ prefixed variables, which prevents command injection and configuration hijacking.
[DYNAMIC_EXECUTION]: Static analysis flagged the use of yaml.load() in scripts/sprint_status.py. However, technical review confirms the script uses the ruamel.yaml library with the RoundTrip loader (typ="rt"), which is the recommended safe method for parsing YAML while preserving comments and formatting. This does not pose the security risks associated with PyYAML's unsafe load() method.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from git commit messages, story files, and session logs. The instructions include strong defensive prompts requiring the agent to ground every finding in a specific source reference (file, line, or commit) and to drop any claim that cannot be verified against the diff. This is a robust mitigation against potential instructions embedded in project metadata.

bmad-retrospective