The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data which could contain malicious instructions designed to manipulate agent behavior.
Ingestion points: Reads project-specific rule files (~/.claude/rules/*.md), memory records (MEMORY.md), and external search results from the gemini_search_grounded tool.
Boundary markers: Absent. The instructions do not specify the use of delimiters or warnings to ignore instructions embedded within the processed files.
Capability inventory: Executes shell commands for updates and audits (claude, npm), and modifies system configuration via the update-config skill.
Sanitization: Absent. There is no logic to filter or escape instructions found in the ingested data.
[COMMAND_EXECUTION]: The skill executes several local CLI commands to perform its audit functions. These include checking the CLI version (claude --version), listing directory contents to find broken symlinks (ls -la), and querying installed MCP servers (claude mcp list).
[EXTERNAL_DOWNLOADS]: The skill performs network operations to check for and apply updates.
NPM Registry: Queries registry.npmjs.org via npm view to find the latest version of the Claude Code CLI.
Marketplace Updates: Refreshes and updates plugins from configured marketplaces using claude plugin marketplace update and claude plugins update commands.
Grounding Search: Uses an external search tool to identify new configuration options in recent releases. These sources are considered well-known or trusted services necessary for the skill's primary function.

setup-check