The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted information from user requests and external web/documentation searches to generate implementation steps in a plan file.
Ingestion points: User-provided task descriptions and external technical documentation retrieved via WebSearch or documentation MCP tools.
Boundary markers: The instructions do not define the use of delimiters or warnings to ignore instructions embedded within the ingested data before it is written to the output file.
Capability inventory: The skill has the ability to write files to the local filesystem (within the .turbo/plans/ directory), execute other skills, and create tasks via TaskCreate.
Sanitization: No explicit sanitization or validation logic is present to filter malicious instructions from the external content before interpolation.
[EXTERNAL_DOWNLOADS]: Fetches information and library documentation using standard WebSearch tools and documentation MCPs to ground technical decisions in current library states.
[COMMAND_EXECUTION]: Executes other internal skills such as /survey-patterns and matching task-specific skills to load relevant guidance into the conversation context.

draft-plan