The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/activity_summary.py executes git via subprocess.run. Findings include:
Use of list-based arguments for subprocess.run instead of shell strings, which prevents standard shell injection.
Proactive validation in ensure_git_repo and parse_git_log functions to reject arguments starting with a dash, mitigating command option injection attacks.
[PROMPT_INJECTION]: The skill processes untrusted commit messages from git logs, representing a potential indirect prompt injection surface.
Ingestion points: Untrusted commit messages are ingested via git log in scripts/activity_summary.py.
Boundary markers: The skill produces structured Markdown and JSON reports, and SKILL.md explicitly instructs agents to focus on the factual components of the report.
Capability inventory: The skill has the capability to execute local git commands via subprocess.run in scripts/activity_summary.py.
Sanitization: Commit messages are reported directly without specific filtering for natural language instructions intended for the LLM.

repo-activity-summary