The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from several external and potentially untrusted sources during its audit process. Ingestion points: It reads all rule files in ~/.claude/rules/*.md, frontmatter from skill directories in ~/.agents/skills/, and hook metadata from hooks.json. Boundary markers: The instructions do not specify any delimiters or explicit 'ignore embedded instructions' warnings for the agent when processing these files. Capability inventory: The skill has high-privilege capabilities including executing shell commands (claude, npm) and modifying system configuration files via the update-config skill. Sanitization: No sanitization, escaping, or validation logic is defined for the content ingested from the filesystem before it enters the agent's context.
[COMMAND_EXECUTION]: The skill executes several shell-based tools to perform its audit, including claude --version, npm view, and claude plugin marketplace update. It also constructs and runs claude plugins update commands using identifiers retrieved from the installed_plugins.json file.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to check for updates and refresh registries. It fetches package metadata for @anthropic-ai/claude-code from the official npm registry and uses the claude CLI to synchronize with external plugin marketplaces.

setup-check