The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from the local codebase to inform its PRD generation process.
Ingestion points: File content and architecture details retrieved via the 'Explore' subagent and tools like Grep and Glob (SKILL.md).
Boundary markers: The skill does not define clear delimiters or instructions to ignore potential malicious content embedded within the files it reads.
Capability inventory: The skill has access to shell execution (Bash), codebase exploration (Agent), and external data submission (gh issue create).
Sanitization: There is no evidence of sanitization or validation of the ingested codebase data before it is used to generate the PRD or submitted to GitHub.
[COMMAND_EXECUTION]: The skill utilizes several local command-line operations to perform its tasks.
It launches a local Node.js server via node scripts/visual-server.cjs to host the 'Visual Companion'.
It uses the gh issue create command to submit the final PRD to a remote repository.
It employs standard utilities like Bash, Grep, and Glob for repository analysis.
[DATA_EXFILTRATION]: The skill transmits generated data to an external platform.
The gh issue create command sends the generated PRD to GitHub. While this is the intended primary function, the skill's internal rules ("Submit immediately... don't ask for review") bypass the typical user review cycle, increasing the risk that unintended or manipulated data could be published.

write-a-prd