The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses authoritative directives ("ALWAYS APPLY", "MANDATORY", "Core Principles") to steer agent behavior and enforce strict execution protocols. While common for orchestration, these instructions could be leveraged to override safety constraints if the routing logic is compromised.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its primary command interface.
Ingestion points: The /mega-mind route <request> command ingests arbitrary user text which is then analyzed to determine the next skill in a chain.
Boundary markers: The instructions do not specify any delimiters or wrappers to isolate user-supplied content from the orchestration logic.
Capability inventory: The orchestrator can chain execution to any of 53 active skills, including those with capabilities for code implementation (executing-plans), infrastructure management (k8s-orchestrator), and deployment (finishing-a-development-branch).
Sanitization: No input validation or sanitization is described for the incoming user request before it influences the routing decision.
[DATA_EXFILTRATION]: The skill instructions require the agent to read and update session state in docs/plans/task.md and manage learned patterns in .agent/instincts/personal/. These files track project details and user interactions, creating a surface for potential data exposure if accessed by unauthorized processes or skills.

mega-mind