The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized access attempts were detected in the skill instructions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest information provided by the user during an interview process to generate new skill files. This attack surface is mitigated by the 'Exit Gate' logic, which requires the agent to read back all content for explicit user approval before any files are written, ensuring a human-in-the-loop validation of the data.
[DYNAMIC_EXECUTION]: The skill generates new markdown-based instructions (skills) in the local .claude/skills/ directory. This is a legitimate functional requirement of the skill for creating project-specific data contexts and is guided by fixed templates and user-confirmed content.
[DATA_EXFILTRATION]: While the skill prompts for database connection details and schemas to document them, it lacks any network-capable commands (such as curl, wget, or fetch) that could be used to exfiltrate this information.

data-context