The Agent Skills Directory

[SAFE]: The skill provides a protocol for 'distilling domain language' through structured questioning. All instructions are focused on this goal and follow standard development practices.
[COMMAND_EXECUTION]: The skill instructs the agent to 'Grep the repo' to find synonym clusters. This involves executing a local search command to analyze existing project documentation and code, which is a common and intended capability for agents operating in a software development context.
[DATA_EXFILTRATION]: No network-enabled tools or patterns (e.g., curl, wget, API calls) were detected. The skill's operations are confined to the local environment.
[PROMPT_INJECTION]: No attempts to override system instructions, bypass safety filters, or extract system prompts were found. The 'ground rules' focus on controlling the flow of the interrogation (e.g., 'One question at a time').
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it reads content from the repository ('Grep the repo').
Ingestion points: The agent reads existing code, CONTEXT.md, and other documentation from the local file system.
Boundary markers: The skill does not explicitly define boundary markers for the data it greps.
Capability inventory: The skill primarily performs text analysis and interrogation. It explicitly delegates state changes (writing to files) to a sibling skill, domain-modeling.
Sanitization: No specific sanitization of grepped content is mentioned, but the constrained nature of the 'grilling protocol' limits the risk of the agent following malicious instructions embedded in the codebase.

distilling-domain-language