The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to orient an agent by resolving and loading content from a unique set of documentation sources (leaf docs, ADRs, and glossary entries). This process creates a significant attack surface where malicious instructions could be embedded in the project's documentation to influence the agent.
Ingestion points: The orientation resolver reads from ARCHITECTURE.md, CONTEXT.md, and various markdown files in the docs/ directory.
Boundary markers: It employs management markers (e.g., ) and unslopify session blocks to track document state.
Capability inventory: The skill has access to the filesystem via fs.readFileSync and executes Git commands for file listing.
Sanitization: While it implements an explicit sanitization protocol for its optional diagnostics record, the documentation content itself is read and processed directly.
[COMMAND_EXECUTION]: The skill includes reference implementations (governance.ts and orientation.ts) that execute the local git binary using execFileSync to perform file enumeration and fingerprinting. These operations are essential for its 'Seam Coherence' checks but involve executing external processes within the host environment.

document-for-agents