The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/naming-ledger.py executes git commands using subprocess.run with argument lists. This approach is secure as it avoids shell interpretation and limits the commands to repository metadata gathering.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository metadata and file structures, which constitutes a potential injection surface. 1. Ingestion points: Git repository file paths, tracked/untracked status, and user-provided dispositions files are parsed by the ledger script. 2. Boundary markers: The instructions explicitly require the tracking ledger to be created outside the repository to prevent contamination of the codebase. 3. Capability inventory: The skill is authorized to move files and update the local ledger file. 4. Sanitization: The helper script performs schema validation and restricts status labels to a predefined whitelist of valid states.

naming-refactor