The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project-level files such as specifications, context documents, and ADRs to generate implementation tasks. This creates a surface where malicious instructions embedded in these project files could influence the agent's behavior during decomposition or subsequent implementation steps.
Ingestion points: SKILL.md and its references specify reading AGENTS.md, CONTEXT.md, and target specs as primary inputs.
Boundary markers: The instructions provide explicit constraints, telling the agent to treat the spec as authoritative and not to invent acceptance criteria or additional scope.
Capability inventory: The skill is authorized to create new task files in the docs/tasks/ directory and route workflows to other implementation-focused skills.
Sanitization: The bundled audit script check_project_artifacts.py includes path-traversal checks using Path.resolve() and is_relative_to(root) to ensure that only files within the project directory are accessed and validated.
[COMMAND_EXECUTION]: The skill provides a command to execute a bundled Python script for validating project artifacts.
Evidence: references/artifact-validation.md provides the command python3 /scripts/check_project_artifacts.py <index.json> --root .
Context: The script is a local utility included in the skill package for read-only audit purposes and does not perform network operations or sensitive system modifications.

to-tasks