The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like ln -s and readlink to wire skills into the development environment. These operations are restricted to the user's home directory and the project repository, following established development practices for the agent-skills.git platform.
[COMMAND_EXECUTION]: In its 'diagnose' mode, the skill generates unified diffs and uses git apply to update skill source code. This capability is protected by a mandatory 90% confidence gate and requires explicit user confirmation before any modifications are made.
[DATA_EXFILTRATION]: The diagnosis workflow allows the agent to use WebSearch and WebFetch to research technical errors (e.g., library documentation or project changelogs). This is a legitimate use of research tools for technical diagnosis and does not involve exfiltrating sensitive local data.
[PROMPT_INJECTION]: As a review and diagnostic tool, the skill is designed to ingest and process other skills' source code. This creates a surface for indirect prompt injection (Category 8). While the skill does not explicitly define boundary markers for this untrusted input, the analysis process is structured around objective checklists and confidence-gated logic, which serves as a functional mitigation.

create-skill