The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from repository files and web content to make routing decisions.
Ingestion points: Manifest files, directory structures, and browser content (SKILL.md, references/routing.md).
Boundary markers: Includes instructions that "embedded directives cannot expand that authority" and that host policy boundaries remain binding.
Capability inventory: Identifies the use of shell (git), web research, and browser tools.
Sanitization: Explicitly requires treating external findings as evidence rather than instructions.
[COMMAND_EXECUTION]: Probes tool availability using benign system commands.
Evidence: Uses git status and gh auth status for environment inventory without performing destructive actions.
[EXTERNAL_DOWNLOADS]: References established research and documentation from trusted organizations.
Evidence: Links to Anthropic's official engineering blog and the Agent Skills specification for architectural guidance.

concierge