The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's Executor and Reviewer agents are instructed to execute build tools, test runners, and linters discovered within the repository's configuration. This creates a capability for the agent to run arbitrary commands defined by the project under development.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions.
Ingestion points: The orchestrator and sub-agents read content from spec.md, plan.md, and review.md, as well as source code files from the repository.
Boundary markers: The skill uses role-based prompts (e.g., agents/reviewers/reviewer-security.md) to define agent behavior, but it does not employ explicit delimiters or 'ignore' instructions when interpolating project files into sub-agent contexts.
Capability inventory: The agents have the capability to write files to the docs/ directory, execute the local scripts/merge-findings.py script, run git commands, and execute project-specific build/test/lint tools.
Sanitization: The merge-findings.py script validates the structure and metadata of findings files but does not sanitize or filter the natural language content of the findings or project specifications for potential injection attacks.

dev-implement-spec