The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to manage the file system and perform automated checks.
It utilizes mkdir -p to create spec folder structures and git mv to move source materials (inbox notes, findings) into reference directories.
It invokes the roundfix CLI tool (roundfix spec check <slug> --stage prd) to validate the generated artifacts before finalizing the report.
[DATA_EXPOSURE]: The instructions require the agent to read specific documentation files that define the project's security and architectural policies.
The agent reads docs/agents/backend.md to identify authentication and HTTP policies.
The agent reads docs/agents/agent-instructions.md to resolve normative clauses and tooling authority constraints.
[PROMPT_INJECTION]: The skill handles untrusted data from an inbox/ directory as input for the PRD creation process.
The risk is mitigated by a strict procedural workflow that requires the agent to inventory, classify, and promote findings before adopting them into the specification folder.

write-prd