The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits external instruction files (AGENTS.md, CLAUDE.md) which may contain untrusted content. 1. Ingestion points: Identified in the 'Discover actual sources' section where the skill reads project and personal instruction files. 2. Boundary markers: The skill includes specific directives to treat inspected files as evidence and to ignore instructions within those files that attempt to override the auditor's task. 3. Capability inventory: The skill has the ability to read and write to the filesystem, including creating backups and applying authorized edits. 4. Sanitization: The instructions require the agent to redact secrets from reports and prevent sensitive content from being exposed in shareable artifacts.
[EXTERNAL_DOWNLOADS]: The skill references official documentation for various AI agent platforms. Evidence: Links to official domains for Anthropic (code.claude.com) and OpenAI (developers.openai.com) are provided for technical guidance.
[COMMAND_EXECUTION]: The skill utilizes command-line tools for searching the filesystem. Evidence: The skill instructs the agent to use 'rg --files' (ripgrep) to discover relevant files within a constrained scope.

agentsmd-optimize