The Agent Skills Directory

[PERSISTENCE]: The skill modifies the target repository's local Git configuration by appending ignore rules to the .git/info/exclude file. It specifically hides the /.first-four-hours/routes/ directory from standard version control checks like git status. This concealment mechanism reduces the visibility of the skill's artifacts and modifications within the project environment.
[COMMAND_EXECUTION]: The skill interacts with the Git CLI through subprocess.run() calls in scripts/route_workspace.py and scripts/seal_route.py. It performs operations such as rev-parse, check-ignore, and ls-files to manage its workspace. These high-privilege commands allow direct manipulation of repository metadata and file status.
[DYNAMIC_EXECUTION]: The execution runtime uses importlib.import_module() to dynamically load diagnostic modules. It also implements a manifest-driven execution model where subprocess.run() is used to launch scripts with arguments constructed from templates and user-provided input, increasing the complexity of the execution flow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests content from an external draft file and persists it in the repository, presenting an attack surface for indirect prompt injection.
Ingestion points: The content of the file specified by the --draft argument in scripts/seal_route.py.
Boundary markers: The script performs structural validation based on markdown headers but does not sanitize the internal prompt instructions.
Capability inventory: The skill has permissions to write files to the repository and modify Git ignore rules.
Sanitization: No sanitization or escaping of the ingested draft content is performed before storage.

router