The Agent Skills Directory

[SAFE]: The skill defines a workflow for managing project documentation, specifically CONTEXT.md for glossaries and Architecture Decision Records (ADRs). It operates strictly within the repository's documentation directory.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it reads and updates user-controlled documentation files.
Ingestion points: The agent reads and processes CONTEXT.md, CONTEXT-MAP.md, and Markdown files located in docs/adr/.
Boundary markers: No specific delimiters or "ignore instructions" prompts are defined for parsing these files.
Capability inventory: The skill allows for reading, creating, and updating Markdown files. It does not possess capabilities for network exfiltration, arbitrary command execution, or sensitive file system access.
Sanitization: The skill does not specify sanitization or validation routines for the content ingested from the documentation files.
Assessment: While the ingestion of untrusted repository data is a known vector, the risk is negligible given the skill's limited capabilities and its intended use as a documentation assistant.

domain-modeling