The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a system for processing external, potentially untrusted data.
Ingestion points: Raw content from URLs and PDFs is fetched and stored verbatim in the external-sources/ folder (SKILL.md).
Boundary markers: The workflow uses frontmatter and structured templates to organize content, though it relies on the referenced platform skill for low-level grounding and linking rules.
Capability inventory: The procedures described involve standard file operations including write, edit, move, and delete (SKILL.md).
Sanitization: The instructions emphasize preserving raw source integrity for traceability; while this creates a surface for instructions embedded in sources to be seen by the agent during research, the skill explicitly mandates human-in-the-loop confirmation for transitions between ingest, research, and consolidation to prevent autonomous misuse.
[SAFE]: References to external resources are limited to official vendor repositories and well-known knowledge-sharing platforms.
The repository link targets the author's official GitHub organization (github.com/inkeep/).
External links include neutral references to public methodology descriptions on GitHub Gist (karpathy).

knowledge-base