The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically executes shell commands, such as validation scripts and verification gates, that are parsed directly from the implementation plan's success-criteria block during Phase 4 (Drive to Green).
[INDIRECT_PROMPT_INJECTION]: The skill processes project plans from the docs/plan/ directory which may contain attacker-controlled content if the repository is untrusted. Malicious instructions or shell commands placed in the success-criteria section of these files could be executed by the agent during the automated build and validation process.
Ingestion points: Implementation plan files located in the docs/plan/ directory as specified in the skill's Phase 0 logic.
Boundary markers: The skill incorporates a manual review checkpoint by prompting the user to confirm the build scope and offering an option to review the plan file before implementation begins.
Capability inventory: The agent has extensive capabilities including Bash command execution, file system modification (writing code and tests), and version control operations such as Git commits, pushing to remotes, and creating pull requests.
Sanitization: The skill does not implement sanitization or validation logic for the verification commands extracted from the plan file before they are executed in the shell environment.

build