The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches a remote 'Master Skill Catalog' from a third-party GitHub repository not associated with a trusted organization or the designated skill author. This content is used to discover and bring other skills into the execution context.
Evidence: SKILL.md contains instructions to fetch the catalog from https://raw.githubusercontent.com/sickn33/agentic-awesome-skills/main/CATALOG.md if local capabilities are insufficient.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it ingests remote, unverified text that directly guides the agent's task-solving strategy.
Ingestion points: Step 3.3 in SKILL.md (Discover and Select Skills) involves fetching and scanning the external CATALOG.md file.
Boundary markers: Absent. The instructions do not define delimiters for the external content or warn the agent to ignore potentially malicious instructions inside the catalog.
Capability inventory: The agent has the capability to search, read, and execute other specialized tools/skills (like @agent-memory-mcp) based on the catalog's content.
Sanitization: Absent. No validation or filtering is applied to the catalog content before the agent uses it for skill selection.

antigravity-skill-orchestrator