The Agent Skills Directory

[SAFE]: The skill's primary function is managing project documentation (Knowledge Base) in the docs/kb/ directory and indexing it in CLAUDE.md. All operations are localized to the file system.
[PROMPT_INJECTION]: The skill proactively includes a security section instructing the agent to never store secrets, API keys, or sensitive credentials, which mitigates accidental data exposure.
[DATA_EXPOSURE]: While the skill reads and writes to the file system, its scope is limited to project documentation files. It does not access sensitive system paths (e.g., SSH keys, AWS credentials).
[REMOTE_CODE_EXECUTION]: There are no patterns of remote code execution, package installation, or dynamic shell commands. The disable-model-invocation flag and lack of network-capable tools further limit risk.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user input and reads back existing KB files. While this creates a potential surface for indirect injection, the risk is minimized by the lack of high-privilege capabilities (like network access) and the requirement for user interaction via AskUserQuestion for critical decisions.

kb-add