The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from task contract files located in production/tasks/. It mitigates injection risks by restricting writes to delimited lifecycle blocks and enforcing path validation. 1. Ingestion points: contract.md files. 2. Boundary markers: delimited lifecycle blocks. 3. Capability inventory: Bash, Write, Edit, Read, Grep. 4. Sanitization: path-based restrictions.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool for administrative file operations and lock management. Security is maintained through strict input validation that ensures task IDs and contract paths are confined to the production/tasks/ directory.
[SAFE]: The skill defines a risk contract that explicitly prohibits high-risk actions such as modifying production code, assets, or version control history, ensuring it only performs authorized administrative updates.

task-done