The Agent Skills Directory

[SAFE]: The skill is designed for codebase analysis and employs standard tools for reading and searching files. The use of a read-only sandbox for complex analysis tasks is a positive security practice.
[INDIRECT_PROMPT_INJECTION]: The skill is inherently exposed to indirect prompt injection due to its primary function of analyzing external codebase content.
Ingestion points: The workflow uses Grep, Glob, and Read tools to ingest code content from the project directory (SKILL.md, Phase 1).
Boundary markers: The prompts generated for the analysis tools (references/prompts.md) do not include specific delimiters or instructions to ignore potentially malicious instructions embedded within the code comments or strings.
Capability inventory: The skill uses Bash(git:*) for version control tasks and mcp__codex__codex for automated exploration. The latter is configured with approval-policy: 'never', allowing the tool to perform its search and read operations without manual intervention.
Sanitization: Ingested code content is passed to the analysis tools without sanitization or filtering of potential prompt injection patterns.

code-investigate