The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/capture-learning.ts executes git rev-parse --show-toplevel to identify the project root for local file storage. This is a standard and safe operation.
[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface (Category 8) by recording narrative text that influences future agent sessions. 1. Ingestion points: Narrative data is accepted from agent/user input in scripts/capture-learning.ts. 2. Boundary markers: Structured with Markdown headers but lacks specific instructions for the agent to ignore embedded commands. 3. Capability inventory: Writes persistent learning logs to .claude/learnings/ which are explicitly intended for later retrieval. 4. Sanitization: Uses a sanitize helper to escape Markdown characters in the global pattern log, which mitigates basic formatting-based injection attacks.

capture-learning