The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires an active network connection to the domain agentic-atlas.dev to access its primary data source and the associated MCP server located at https://agentic-atlas.dev/mcp/.
[PROMPT_INJECTION]: The instructions include specific behavioral constraints (e.g., "Never answer from memory," "Status are honest at every tier") and rigorous formatting requirements for subagents. While these are defensive in nature, they represent direct control over the agent's internal reasoning logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes content fetched from an external source, which introduces a potential vector for indirect prompt injection.
Ingestion points: External data enters the agent context via the atlas_orient, atlas_cards, atlas_read, atlas_navigate, and atlas_define tools in SKILL.md.
Boundary markers: The skill explicitly defines a "contract" for the subagent's output, requiring specific Markdown bullet patterns and citations to help the agent distinguish instructions from retrieved data.
Capability inventory: The skill is primarily focused on information retrieval and routing; it does not define capabilities for file writing, subprocess execution, or administrative system modifications.
Sanitization: The skill mandates a validation step where the agent must check the structure of returned content against the predefined contract and discard malformed results.

consulting-patterns