The Agent Skills Directory

[PROMPT_INJECTION]: The skill displays a vulnerability surface for indirect prompt injection by design.
Ingestion points: Untrusted data enters the agent context through user-provided arguments in SKILL.md and through generated plan files (e.g., zhongshu-plan.md and dispatch-plan.md) which are read and processed by the coordinator, menxia, and shangshu roles.
Boundary markers: The skill uses basic markdown formatting and headers, but lacks explicit instruction delimiters or 'ignore instructions' guards when interpolating potentially attacker-controlled content into prompts for sub-agents.
Capability inventory: The skill possesses extensive capabilities, including spawning new agents (Agent), writing and editing files (Write, Edit), and executing system commands (Bash) and analysis tasks via ccw cli.
Sanitization: There is no evidence of content validation or escaping before external data is interpolated into prompt templates for subsequent agent phases.

team-edict