The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local administrative scripts using the Node.js runtime to manage project state. Specifically, it runs .agents/scripts/precheck.mjs for environment validation and .agents/scripts/cooking.mjs to track and update the status of feature development units.
[INDIRECT_PROMPT_INJECTION]: The orchestration logic interpolates untrusted data from users and other sub-agents into subsequent task descriptions, which could allow malicious instructions to influence the agent's behavior.
Ingestion points: Untrusted data enters the agent context through user-provided requirements, the goal.md file content, and the <原文> placeholder which contains raw report text from previously executed sub-agents (e.g., in the sync-docs and to-tasks prompts).
Boundary markers: The prompt templates in references/subagent-prompts.md do not utilize strong delimiters or explicit "ignore embedded instructions" warnings for the interpolated <原文> or <feature> data.
Capability inventory: The skill possesses the capability to execute shell commands via local Node.js scripts and perform automated version control operations via git-commit.
Sanitization: There is no evidence of sanitization, escaping, or schema validation for the data being passed between agents; raw text is directly interpolated into prompt strings.

rush