The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill ingests data from the local codebase and external research sources to generate a plan that it instructs the user to later execute. This creates a surface where malicious instructions embedded in those external sources could be incorporated into the generated plan.
Ingestion points: Codebase files during Phase 2 (Codebase Intelligence) and documentation/API research in Phase 3.
Boundary markers: Absent. The instructions do not define delimiters or provide guidance to disregard commands or instructions found within the analyzed content.
Capability inventory: Use of the Write/Edit tool to create plan files and the prompt to the user to run the /execute command on the output.
Sanitization: Absent. No filtering or validation of ingested content is performed before it is written into the plan file.
[COMMAND_EXECUTION]: Local File Verification. The skill executes shell commands to confirm the existence and size of the output plan file.
Evidence: Shell execution of test -f .agents/plans/[feature-name].md and wc -l .agents/plans/[feature-name].md during the verification phase.

plan-feature