The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/git-activity.sh performs local operations using git, jq, and date. It accesses user-specified directories to retrieve commit history.
[INDIRECT_PROMPT_INJECTION]: The skill ingests git commit messages (external data) and stores them in JSON format. This creates a surface for indirect prompt injection where malicious instructions inside commit messages could affect subsequent AI analysis.
Ingestion points: The skill uses git log in scripts/git-activity.sh to extract subjects and bodies of commits.
Boundary markers: Output is stored in structured log.json files.
Capability inventory: The skill can read local repository history and write to the filesystem at ~/.cache/recap/git/.
Sanitization: Content is escaped for JSON using jq, but no semantic filtering is applied to the commit text.

git-activity