The Agent Skills Directory

[COMMAND_EXECUTION]: The core instructions in SKILL.md direct the agent to use filesystem tools like Test-Path to inspect project artifacts (e.g., package.json, Cargo.toml, docs/prd*.md). This is a legitimate function for an orchestrator to determine the current state of a project.
[PROMPT_INJECTION]: Multiple reference files (including architecture-patterns.md, state-management.md, and security-best-practices.md) each contain 150 repetitive sections of technical-sounding filler text. This massive volume of redundant data (approximately 1,200 sections across 8 files) is a form of 'token flooding' or 'context stuffing.' This technique is often used in adversarial contexts to overwhelm an LLM's context window, which can cause the model to ignore earlier system instructions or safety guardrails.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It is designed to ingest and process data from untrusted local files (project manifests and documentation) to determine routing logic. It lacks explicit boundary markers or instructions to ignore potential commands embedded within those external files, creating a low-severity path for indirect control of the agent's routing behavior.

master-orchestrator