The Agent Skills Directory

[SAFE]: The skill instructions and metadata describe a legitimate workflow for technical documentation and RFC management within a controlled project environment.\n- [COMMAND_EXECUTION]: The skill utilizes shell commands such as ls, grep, and cat via an exec tool to manage files. These operations are scoped to the proposals/ and decisions/ directories to allow the agent to sequence documents, search for keywords, and read existing records as part of the software design lifecycle.\n- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests data from local markdown files. Evidence:\n
Ingestion points: Content from existing proposals and decisions read via grep and cat in SKILL.md.\n
Boundary markers: The workflow includes a mandatory human-in-the-loop 'STOP gate' (Step 1) requiring the user to validate the problem framing before the agent proceeds to generate a design, which serves as a mitigation against adversarial content in ingested files.\n
Capability inventory: The skill has the ability to write and edit files and exec local shell commands.\n
Sanitization: No explicit escaping or filtering for ingested markdown content is specified in the instructions.

frame-a-proposal