SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillscat-kit-agent-context

cat-kit-agent-context

Agents & models

View repository

Updated 10-16 08:34

About this skill

@cat-kit/agent-context

Install

npm add @cat-kit/agent-context

Verify API

generated/ under .d.ts consistent with npm dist ; ESM JavaScript files in the release artifact uniformly use the .js suffix, and the scripts required for Skill installation are published in dist/skill/scripts/。

Runtime environment: Node.js (CLI / toolchain).

Skill artifact conventions

agent-context install generated ac-workflow aligns with Agent Skills progressive disclosure:

  • By default, only .agents/skills/ac-workflow/ is rendered as the canonical source;--tools only creates an optional compatibility entry point
  • The compatibility entry point preferentially symlinks / junctions to the canonical source; when symlinks are unsupported or a regular directory already exists, it syncs via copy fallback
  • SKILL.md is a short navigation entry point:onestartup command (node <SKILL_DIR>/scripts/get-context-info.js, which also performs format validation), looks up the routing table by status + user intent, then reads the protocol
  • The full protocol text is located in references/*.md, and after the action is determined, only the corresponding file is read
  • references/ask-user-question.md contains the red line of "when to prohibit asking questions" and the reverse follow-up questioning method; the protocol does not hardcode host tool names
  • references/_principles.md is the shared professional competency baseline for the planning / implementation / review roles; the protocol only references it without repeating it
  • Context status and numbering come from scripts/get-context-info.js output (JSON + exit code); agentsare prohibited from scanning directories on their own to infer
  • scripts/validate-context.js serves only as an independent validation fallback when the main script fails to start
  • description needs to avoid accidental triggering by ordinary coding, code review, planning, or mere AGENTS.md modifications
  • The review follow-up at the end of the protocol is triggered by complexity thresholds (scope of impact, external APIs, databases, security-sensitive, etc.), and is no longer asked every time
  • rush scenarios are prohibited from triggering mandatory reverse follow-up questioning; if a task requires reverse follow-up questioning, switch to plan
  • agent-context context / skill-eval provides a CLI-equivalent integration point;prompt-gen generates a general template by default; personal preferences go through --profile whj

More