The Agent Skills Directory

[PROMPT_INJECTION]: The orchestration logic ingests external data from specification documents (PRDs, Notion) and repository source code, which is then interpolated into prompts for specialized sub-agents. This creates a surface for indirect prompt injection where malicious instructions in a spec file could influence sub-agent behavior.
Ingestion points: specification files and repository code provided by the user.
Boundary markers: prompt templates use markdown headers to organize context but lack explicit delimiters to isolate untrusted content from instructions.
Capability inventory: the system can spawn multiple agents with repository access and execution tools.
Sanitization: external text is used in templates without explicit sanitization or filtering.
[COMMAND_EXECUTION]: The skill facilitates software implementation and testing by spawning sub-agents that use planning and execution tools (e.g., superpowers:executing-plans). This functionality is governed by a 'Monorepo Rule' that restricts each agent to a single repository and a 'Validation Gate' that requires git-based checks of all modifications before delivery.
[SAFE]: The skill implements professional state management in a dedicated .state/ directory with specific instructions to exclude this data from version control via .gitignore. It also utilizes a structured wave system to ensure that architecture and testing phases occur before any code implementation, providing layers of human-in-the-loop and peer-agent review.

teamlead-subagent