The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface.
Ingestion points: Untrusted task descriptions and user-provided decisions are ingested as part of the routing logic in SKILL.md and incorporated into the CONTEXT field of the references/brief-template.md.
Boundary markers: The skill utilizes a structured format in references/brief-template.md (using headers like GOAL, SCOPE, and CONTEXT) to separate different types of information, which provides basic structural separation but lacks explicit 'ignore embedded instructions' warnings.
Capability inventory: The lead agent has the capability to spawn and manage sub-agents using the pi command and the herdr skill, as described in references/conventions.md.
Sanitization: There are no explicit instructions or automated steps defined to sanitize or escape data before it is passed from the lead agent into the execution environment or briefs of sub-agents.
[COMMAND_EXECUTION]: Sub-agent process management.
Evidence: references/conventions.md instructs the agent to use the pi tool (e.g., "Start Pi as a persistent interactive process") to initiate and manage delegated work streams.
[NO_CODE]: The skill consists of documentation and instructional templates.
Evidence: The analyzed files (SKILL.md, references/brief-template.md, and references/conventions.md) contain markdown-based instructions and templates rather than executable code files like Python or JavaScript.

teamlead-mode