The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill programmatically generates TypeScript files (hooks/*.ts) and learned skills based on user-provided feedback and historical session transcripts. This runtime creation of executable content allows untrusted historical data to define logic that is subsequently executed by the environment.
[PROMPT_INJECTION]: This skill automates the modification of other skills' behavioral constraints (Iron Laws and Red Flags) using data ingested from session history. This creates an indirect prompt injection vector where malicious instructions embedded in past conversations could be promoted to persistent system-level rules. Evidence Chain: 1. Ingestion points: MEMORY.md, <memory_dir>, CLAUDE_TRANSCRIPT_PATH. 2. Boundary markers: Absent. 3. Capability inventory: Write, Edit, Bash (grep), skill-creator delegation. 4. Sanitization: Absent.
[COMMAND_EXECUTION]: The skill uses the grep utility to search through local memory files and transcripts using patterns derived from user corrections and keywords.

pattern-capture