The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator command and its worker agents utilize shell commands (including find, grep, git log, and wc) to perform project-scale analysis, contributor archaeology, and pattern searching within the codebase as part of the orientation and risk assessment phases.
[DATA_EXFILTRATION]: The skill explicitly instructs agents to search for and extract sensitive information, such as environment variables containing KEY or SECRET, authentication middleware configurations, and credential-bearing files like .env, to produce its 'Risk Assessment' and 'Context Diagram' artifacts. While this is consistent with the skill's stated purpose of auditing and documentation, it represents an inherent risk for sensitive data exposure.
[PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection because it ingests entire codebases (untrusted data) and possesses powerful capabilities such as file system modification and the ability to spawn autonomous agents.
Ingestion points: Multiple agents (e.g., system-overview, component-map) read the project root and all source files during the scanning process.
Boundary markers: No explicit boundary markers or 'ignore' instructions are used when passing file content into sub-agent prompts.
Capability inventory: Access to Read, Write, Edit, Bash, and Agent tools across all 16 orchestrated workers.
Sanitization: No sanitization or escaping of ingested codebase content is performed before interpolation into agent instructions.

codebase-docs-expert