The Agent Skills Directory

[DATA_EXPOSURE]: The skill records execution metadata, such as task duration and modified file paths, in a local 'Chronicle' log. This system is designed for operational monitoring and stores data exclusively within the project's local memory directory.
[DATA_EXPOSURE]: An opt-in 'Trajectory Capture' feature records session details for evaluation. It includes a built-in redaction process that scans for and masks potential secrets, including API keys and authentication tokens, before any data is written to disk.
[INDIRECT_PROMPT_INJECTION]: The skill performs 'Feed-Forward' and 'Mutation Analysis' by ingesting historical logs and retrospectives. Ingestion points: project-local 'patterns.md', 'signals.jsonl', and retrospective files. Boundary markers: Instructions and historical data are separated into specific sections within analysis prompts. Capability inventory: The skill is limited to writing proposal files and dispatching internal subagents. Sanitization: It utilizes an automated redaction pipeline and maintains a strict 'human-in-the-loop' requirement for any resulting skill changes.
[COMMAND_EXECUTION]: The skill manages analytical tasks by dispatching subagents using pre-defined templates. These operations occur within the agent's controlled environment using standard tools.

forge