The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from session transcripts and memory files to generate new enforcement mechanisms, creating a risk where malicious instructions in past history are turned into permanent rules.
Ingestion points: Processes CLAUDE_TRANSCRIPT_PATH, memory files in <memory_dir>, and direct user feedback.
Boundary markers: Lacks explicit delimiters or instructions to ignore embedded commands when processing ingested content.
Capability inventory: Modifies SKILL.md files to add Iron Laws and Red Flags, creates new memory entries, and generates TypeScript validation hooks.
Sanitization: Content from transcripts is not escaped or sanitized before interpolation into generated artifacts.
[DYNAMIC_EXECUTION]: The skill generates TypeScript code for validation hooks and writes them to the hooks/ directory for subsequent execution by the agent platform.
[COMMAND_EXECUTION]: Uses the grep utility to search local files, using patterns that could be influenced by untrusted user input.

pattern-capture