The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands including git log, git diff, and ls to inspect session history and locate existing documentation. These operations are restricted to identifying changes and files within the project and agent configuration directories.\n- [DATA_EXFILTRATION]: Accesses and modifies files within the user's home directory (~/.claude/, ~/CLAUDE.md) and the project root. This behavior is necessary for the skill's primary function of maintaining cross-session memory and does not involve network communication or unauthorized data exposure.\n- [PROMPT_INJECTION]: The skill identifies and persists session outcomes, creating a surface for indirect prompt injection where malicious instructions in a session transcript could be saved to durable memory.\n
Ingestion points: Ingests data from session transcripts and git logs to identify lessons and decisions.\n
Boundary markers: The workflow follows structured steps to categorize data, though it lacks explicit delimiters for the raw transcript content.\n
Capability inventory: Includes file system read/write access and local command execution tools.\n
Sanitization: Employs an 'Anti-patterns' section instructing the agent to focus on validated outcomes and avoid routine or unconfirmed entries.

handoff