Claude Code¶

Pipeline concepts from this training are IDE-agnostic. If you use Claude Code instead of Cursor, you can adapt the setup. This page consolidates all Claude Code references from across the repo.

Concept mapping table¶

Cursor	Claude Code
`.cursor/rules/*.mdc`	`CLAUDE.md` at repo root
`.cursor/agents/*.md`	Instructions referenced in `CLAUDE.md`
`.cursor/skills/*/SKILL.md`	`.claude/commands/*.md`
`subagent_type` dispatch	Sequential prompting through stages

What transfers directly¶

Artifacts, pipeline concepts, agent roles, and stage gates are IDE-agnostic. The same plan.json, worker-result.json, test-result.json, review-result.json, debug-diagnosis.json, and git-result.json work in both environments. Walkthrough content and pipeline artifacts are portable.

What differs¶

Aspect	Cursor	Claude Code
Wiring mechanism	`subagent_type` dispatch — Task tool invokes specialized subagents	Sequential prompting — you prompt through stages in order
Rule activation	Frontmatter + `description`; Cursor decides when to include rules	`CLAUDE.md` directives; you structure instructions in one file
Skill discovery	`.cursor/skills/*/SKILL.md`; Cursor surfaces skills to agents	`.claude/commands/*.md`; you reference commands explicitly
Tiered routing	`jg-tier-routing.mdc` + subagent dispatch with tier suffixes	Routing logic in `CLAUDE.md`; model selection per prompt

Migration path¶

To adapt the .cursor/ setup for Claude Code:

Convert rules to CLAUDE.md sections — Extract the body of each .mdc rule and add it as a section in CLAUDE.md. Include the jg-planner-first gate, tier routing logic, and any team rules.
Convert skills to .claude/commands/ — Each SKILL.md becomes a command file. The jg-pipeline-artifact-io skill (artifact layout, read/write conventions) maps to a command that agents invoke when reading or writing pipeline artifacts.
Use sequential prompting instead of subagent dispatch — Instead of Task(subagent_type="jg-subplanner", ...), prompt: "You are the subplanner. Read the issue and produce plan.json at .pipeline//plan.json." Then: "You are the worker. Read plan.json and implement..." and so on. Model selection per stage replaces tier-based subagent selection.
Preserve artifact schemas — The pipeline/schema.py and artifact shapes in pipeline/README.md apply unchanged. Validation and stage-gate checks work the same.