The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions to bypass or override platform-level modes, such as Claude Code's plan mode, and to ignore specific instructions within sub-skills to maintain the orchestrator's state.
[COMMAND_EXECUTION]: The skill uses command-line tools like gh issue view to fetch issue data and remotion or Playwright to perform visual quality assurance on the implemented code.
[DATA_EXFILTRATION]: The skill makes network requests to GitHub to fetch issue content. While GitHub is a well-known service, this ingestion of external data without proper boundaries is a security consideration.
[PROMPT_INJECTION]: (Indirect Prompt Injection Surface)
Ingestion points: Data is fetched from GitHub issue bodies via gh issue view and read from local project specification files.
Boundary markers: The instructions do not provide delimiters or specific guidance to the agent to disregard instructions potentially embedded within the fetched external text.
Capability inventory: The agent is authorized to write to the file system, execute sub-agents, and run developer tools throughout the execution phases.
Sanitization: There is no evidence of validation or sanitization of the content retrieved from GitHub before it is processed as a task description.

forge-plan