The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external task specifications and plans that drive the behavior of developer and verifier sub-agents.
Ingestion points: Input specifications (e.g., spec.md, delta-*.md) and task descriptions provided via the /bro-do-it command.
Boundary markers: The skill orchestrates tasks using specific templates (e.g., developer-prompt.md) and instructs the agent to treat plans as 'immutable sources of truth', though it relies on the model's instructions to resist adversarial content within those inputs.
Capability inventory: The skill allows file system modification (implementation), shell command execution (tests, builds), and local git operations (commits).
Sanitization: Safety is managed through 'Implementation Hygiene' rules (defined in references/implementation-hygiene.md) which explicitly forbid storing secrets, modifying live databases, or bypassing established project mechanisms.
[COMMAND_EXECUTION]: The verification process involves executing shell commands to validate the implementation.
Evidence: subagents/verifier-prompt.md instructs the sub-agent to 'Run relevant automated tests, standard static checks, type checks, and build.'
[DYNAMIC_EXECUTION]: The skill facilitates the execution of project-specific scripts and tests to verify development tasks.
Evidence: The verifier sub-agent is required to execute code within the project environment to provide 'reproducible proof' of task completion.

bro-do-it