The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted natural language descriptions of applications to generate structured data models. This represents an indirect prompt injection surface where malicious input could attempt to influence the generated output. This is mitigated by a mandatory structural approval gate (Phase A) where the user must explicitly review and approve the ASCII resource map before the agent is permitted to write the Phase B artifacts (template.json and template.md) to disk. The instructions also enforce strict mapping to Form.io primitives, limiting the impact of malicious data.
Ingestion points: User-provided application descriptions (SKILL.md).
Boundary markers: Structural separation between the interview/planning phase and the file emission phase.
Capability inventory: write_file (template.md, template.json), ask_user_question (SKILL.md).
Sanitization: Instructions ground all output in strict Form.io constructs (resources, forms, actions, components).
[DATA_EXPOSURE]: The skill is designed to write architectural intent documents and project templates to the local filesystem using the Write tool. This is the primary intended function of the skill and does not involve accessing sensitive system files or environment variables.
[COMMAND_EXECUTION]: The repository includes a grade.py Python script within the evals/ directory. This script is intended for developers to run locally to measure the skill's performance against test cases. It is not invoked or executed by the AI agent during typical interactions with an end-user.

formio-resource-planner