The Agent Skills Directory

Indirect Prompt Injection (MEDIUM): The skill ingests untrusted data from GitHub commit messages and local journal files. An attacker could craft a commit message that, when summarized, influences the agent's behavior during the review process. \n
Ingestion points: GitHub commit logs (via gh tool), journal entries in my-vault/, and project worklogs. \n
Boundary markers: Absent. The skill does not use delimiters or instructions to ignore embedded commands in the data it reads. \n
Capability inventory: Write, Edit, and restricted Bash. The agent can modify notes and memory files based on what it reads. \n
Sanitization: None. Data is read and summarized directly into the context.\n- Command Execution (LOW): The skill utilizes the Bash tool to perform date calculations and fetch GitHub data. Security risk is mitigated by the allowed-tools metadata which restricts bash usage to the gh and date binaries only, preventing arbitrary shell command injection.

weekly-review