The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large volumes of untrusted, historical conversation data from ~/.claude/projects/ and life-os/Obsidian/ to analyze user behavior.
Ingestion points: Processes all .jsonl files in ~/.claude/projects/, all memory directories, and the user's entire Obsidian vault (life-os/Obsidian/) including stream-of-consciousness transcriptions and personal goals.
Boundary markers: The instructions do not explicitly mandate delimiters or escaping when interpolating this historical data into the analysis prompt, though it does use a custom Python script to filter out tool noise.
Capability inventory: The skill uses scripts/extract_conversations.py to read and write files, and instructions permit saving the final report back into the file system (Obsidian/unprocessed/).
Sanitization: The extract_conversations.py script performs basic filtering of known 'noise' prefixes but does not sanitize the remaining human-authored text for embedded instructions or prompt injections.

metacognitive-architect