The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on 'recipes' (Markdown files like test-execution.md and review.md) which define the exact commands used for testing and linting. This constitutes an indirect prompt injection surface.
Ingestion points: Recipes are loaded from the project's own repository (<projectPath>/process-recipes/) or a catalog service via the catalog-identifier sub-agent.
Boundary markers: The agent is explicitly instructed not to read the content of these recipes directly. Instead, it passes the paths to specialized scripts (implement-actions.sh) which handle the parsing and execution.
Capability inventory: The skill uses the Bash tool to execute the commands defined in the recipes during the build, fix, and finish stages. It also dispatches various sub-agents (implementer, reviewer, fixer) to perform specific tasks.
Sanitization: The instructions specify that the internal scripts expand placeholders like {paths} and {a.b} internally to prevent shell injection, and they use 'read' and 'write' hooks to enforce access controls during the implementation process.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute implement-actions.sh and task-actions.sh. These scripts are part of the skill's own package and are used to manage the implementation lifecycle, including environment setup and commit management. This is the intended primary purpose of the skill.
[DYNAMIC_EXECUTION]: The skill supports 'recipes' that can define custom scripts or configuration gates. These are written into the worktree at runtime, executed, and then removed. This is a controlled feature for project-specific automation and is managed by the skill's own internal logic.

implement