The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates the 'obsidian' skill to search and create files within a local vault path. This is consistent with its stated purpose of managing architecture records.
[DATA_EXPOSURE]: The skill reads from local project documentation such as 'CLAUDE.md', 'LEARNING.md', and 'MEMORY.md' to gather context. It also defines a fallback local file path for writing. All operations are confined to the local filesystem with no network exfiltration detected.
[PROMPT_INJECTION]: An indirect prompt injection surface exists as the skill processes untrusted project data. Evidence: 1. Ingestion: Step 2 researches the codebase and learning logs; 2. Boundaries: No specific delimiters are defined in instructions; 3. Capabilities: obsidian tool execution and file system writes; 4. Sanitization: No sanitization steps are listed. This risk is managed by the skill's structured format and internal review steps.

adr