The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from change artifacts such as feature files, proposals, and design documents, which are then passed to independent subagents.
Ingestion points: Reads files from beat/changes/<name>/ (including features/*.feature, proposal.md, design.md, and tasks.md).
Boundary markers: The main skill logic passes these artifacts as data inputs to the Agent tool for subagent execution.
Capability inventory: The skill can execute local shell commands for testing suites and perform file writes to status.yaml.
Sanitization: No explicit sanitization or filtering of the content from these artifacts is mentioned before they are interpolated into subagent prompts.
[COMMAND_EXECUTION]: The skill is designed to execute local shell commands as part of its primary verification workflow.
Evidence: SKILL.md explicitly calls for running automated tests using detected frameworks (e.g., testing.behavior and testing.e2e). Additionally, code-reviewer-prompt.md instructs the subagent to run git diff commands to review implementation changes.

verify