The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator and its sub-agents utilize shell commands for project environment setup, version control operations (git), and automated code validation. These actions include creating directories, executing linters, and running test suites (e.g., pytest, npm test), which are standard and necessary for a software development lifecycle tool. \n- [PROMPT_INJECTION]: The skill ingests external data from Product Requirement Documents (PRDs) and codebase analysis reports to generate Project Requirement Plans (PRPs). This architecture presents an indirect prompt injection surface where malicious instructions in a PRD could influence the agent's behavior. However, the workflow incorporates architectural safeguards, such as spawning fresh sub-agent sessions for each task and employing an independent 'Validator' agent to verify that the implementation matches the original requirements. \n
Ingestion points: PRD files and codebase research content processed by the generate-prp and execute-prp modules. \n
Boundary markers: Tasks are encapsulated in specialized sub-agent sessions with dedicated instruction files (piv-executor.md, piv-validator.md). \n
Capability inventory: The skill has the capability to write files, execute shell commands, and manage git repositories within the project scope. \n
Sanitization: The validation step acts as a functional check against the implementation, though explicit prompt sanitization logic is not detailed. \n- [SAFE]: No evidence of credential theft, hardcoded secrets, persistence mechanisms, or malicious obfuscation was found. The skill identifies itself clearly and directs its metadata and commit messages to a public GitHub repository associated with the tool.

piv