The Agent Skills Directory

[SAFE]: The skill performs legitimate writing assistance functions by detecting keywords and loading corresponding local documentation from the .claude/knowledge-base directory. No obfuscation, malicious instructions, or data exfiltration attempts were detected.\n- [INDIRECT_PROMPT_INJECTION]: The skill reads external project documentation files based on user-triggered keywords (e.g., reading .claude/knowledge-base/requirements/{name}.md). While this creates an ingestion point for potentially untrusted project data, the behavior is localized to the user's project and is the central intended feature of the tool, presenting no inherent threat to the agent's safety guidelines.\n- [COMMAND_EXECUTION]: The skill uses the Edit tool to write configuration to specification.md. This is a standard project management operation and is correctly scoped to the allowed tools provided in the frontmatter.

requirement-detector