The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the codebase (source code, comments, documentation) using Read, Grep, Glob, and gitnexus tools. This data is interpolated into prompts for explorer and explainer subagents without sufficient isolation or sanitization.
Ingestion points: SKILL.md (via gitnexus), references/explorer-prompt.md (via Read/Grep/Glob), and references/explainer-prompt.md (via explorer findings synthesis).
Boundary markers: Absent. The prompt templates do not use specific delimiters or instructions to ignore instructions embedded in the source code files.
Capability inventory: The skill executes the gitnexus CLI and uses subagents (configured as readonly: true) to perform broad repository searches and reads.
Sanitization: No escaping or validation is performed on the content read from files before it is injected into the next agent's context.
[COMMAND_EXECUTION]: The skill executes the gitnexus CLI tool if it is present on the system PATH. It runs gitnexus analyze --index-only to refresh the index and various query commands (gitnexus query, gitnexus context, gitnexus trace) to gather information about the repository structure.

how