The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes untrusted data (changed files, git diffs) to update project guidelines in CLAUDE.md.
Ingestion points: Changed files, content summaries, and Git context (commits, diffs).
Boundary markers: Absent; the skill does not specify delimiters to separate untrusted file content from processing logic.
Capability inventory: Reads files, uses grep for codebase searching, and writes to CLAUDE.md.
Sanitization: Absent; no explicit validation or escaping of ingested file content is mentioned before it is committed to documentation.
Command Execution (SAFE): The skill describes using grep and observing previous bash executions. These are standard operations for a developer tool and are used here for documentation verification rather than executing arbitrary attacker-controlled strings.

memory-processor