The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Node.js scripts (scripts/check-plan.mjs and scripts/consumers.mjs) to validate project documentation and search for code consumers.
[COMMAND_EXECUTION]: The consumers.mjs script utilizes git CLI tools (git rev-parse and git grep) to identify code references associated with changes in technical specifications. The use of the -F (fixed strings) flag in git grep ensures that search patterns are treated literally, mitigating regex-based injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided or project-specific Markdown files (plan.md, spec.md, handoff-pack). While these files represent an ingestion point for external data, the skill uses strict template-based validation and structural parsing, which limits the risk of adversarial content influencing the agent's core logic.
Ingestion points: Processes plan.md, spec.md, and handoff packets from the local filesystem or standard input.
Boundary markers: Uses specific Markdown headers (e.g., ## 进度总览, ## 责任分级) and table headers as structural boundaries.
Capability inventory: File system read/write access via node:fs and shell command execution via node:child_process (limited to git).
Sanitization: Implements fixed-string searching for codebase queries and structural validation against predefined templates.

spec-and-plan