The Agent Skills Directory

[COMMAND_EXECUTION]: Executes shell commands via Bash to perform system-wide health checks, including file validation, manifest parsing, and hook registration audits. This is consistent with its primary purpose as a diagnostic utility.
[PROMPT_INJECTION]: Identified an indirect prompt injection surface because the skill reads and potentially displays data from untrusted repository files.
Ingestion points: Reads frontmatter from SKILL.md files, content from manifests/*.json, and the .claude/memory/decisions.jsonl log.
Boundary markers: Absent; the skill does not use specific delimiters to isolate external file content from instructions.
Capability inventory: The skill has access to Bash, Read, Grep, and Glob tools.
Sanitization: Uses Python's json module for integrity validation, but error reporting may output raw, unescaped content from corrupt lines, which could contain malicious instructions.
[EXTERNAL_DOWNLOADS]: Recommends the installation of optional dependencies from external sources, including the trusted vercel-labs organization for the agent-browser skill and the agentation-mcp package. These are documented as configuration checks for missing components.
[DATA_EXPOSURE]: Inspects the environment for the presence of sensitive variables such as TAVILY_API_KEY to verify MCP server configuration. The skill validates the existence of these keys without displaying, logging, or transmitting their values.

doctor