The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The agents/skills/00-index.md file references third-party community collections and marketplace websites, such as skillsmp.com and skillhub.club, as well as several unverified GitHub repositories (karanb192/awesome-claude-skills, VoltAgent/awesome-agent-skills, hesreallyhim/awesome-claude-code).
[REMOTE_CODE_EXECUTION]: The registry provides instructions for installing and running external code via package managers (npm install -g namnam-skills) and specialized CLI tools (npx skills install), which can lead to the execution of unvetted third-party content.
[DYNAMIC_EXECUTION]: The skill is designed to discover and potentially "auto-adopt" new or updated capabilities found in the environment. While it includes a validation mandate in its security section, this dynamic discovery loop introduces a mechanism for loading external logic at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and summarizes content from index files and project memory, which could be manipulated to influence agent behavior.
Ingestion points: Reads agents/skills/00-index.md, agents/guides/00-index.md, and agents/memory/PROJECT_STATE.md to build its report.
Boundary markers: Uses a structured SESSION_PREFLIGHT report format and explicit health-check phases to separate initialization from execution.
Capability inventory: Primarily file-read operations and delegation to sub-skills for state restoration and updates.
Sanitization: Includes explicit instructions to treat non-repo assets as untrusted and requires validation (e.g., egress scans) before integration.

preflight