Skip to content

Agent Registry

All agents across Foundation, Practitioner, and Expert tiers. Foundation uses 3 agents for learning; Practitioner adds the full pipeline; Expert adds tiered variants for complexity-based routing.

Full Agent Table

Agent Model Tier(s) Role Reads Writes
jg-planner gemini-3.1-pro Orchestrator; classifies complexity, routes pipeline and failures All artifacts (read-only), issue state.yaml (optional)
jg-subplanner gpt-5.1-codex-max Standard Decompose issue → ordered plan Issue body/comments plan.json
jg-subplanner-high gpt-5.1-codex-max High Decompose with dependency graphs and risk analysis Issue body plan.json
jg-worker-fast gemini-3-flash Fast Single-file edits; escalates if exceeds scope plan.json worker-result.json
jg-worker gpt-5.3-codex Standard Implement and test per plan plan.json, debug-diagnosis.json worker-result.json
jg-worker-high gpt-5.1-codex-max High Complex implementation with risk assessment plan.json, debug-diagnosis.json worker-result.json
jg-tester-fast gemini-3-flash Fast Phase 1 only (lint, typecheck, unit tests) (runs commands) test-result.json
jg-tester gemini-3-flash Standard Phase 1 + Phase 2 verification (runs commands) test-result.json
jg-reviewer-fast gemini-3-flash Fast Scope check and lint-level review plan.json review-result.json
jg-reviewer gemini-3.1-pro Standard Quality gate; scope and slop plan.json, worker-result.json review-result.json
jg-reviewer-high gemini-3.1-pro High Deep review with architecture and security analysis plan.json, worker-result.json review-result.json
jg-debugger claude-4.6-sonnet Standard Classify and diagnose failures test-result.json, plan.json debug-diagnosis.json
jg-debugger-high claude-opus-4.6 High Multi-causal analysis, cross-module tracing test-result.json, plan.json debug-diagnosis.json
jg-git gemini-3-flash Branch, commit, PR (no merge) (git state) git-result.json
jg-benchmarker gemini-3-flash Pull benchmarks; evaluate cost/performance; recommend models per agent Benchmark sources, snapshot schema benchmarks/snapshots
team-linter gemini-3-flash Runs project linter and writes lint result plan.json, worker-result.json lint-result.json

jg-benchmarker is a support agent (on-demand), not a pipeline stage. team-linter is project-specific (add via tutorial or copy from sandbox).

Tier Routing Matrix

Complexity Subplanner Worker Tester Reviewer Debugger
Trivial (skip) jg-worker-fast jg-tester-fast jg-reviewer-fast (skip)
Standard jg-subplanner jg-worker jg-tester jg-reviewer jg-debugger
Complex jg-subplanner-high jg-worker-high jg-tester jg-reviewer-high jg-debugger-high

Pipeline Execution Order

  1. jg-planner — Entry point. Classifies complexity; invokes subplanner (or gives worker direct scope for trivial).
  2. jg-subplanner[-high] — Writes plan.json (affected_files, steps, acceptance_mapping).
  3. jg-worker[-fast|-high] — Implements; writes worker-result.json. 3.5. team-linter (optional) — After worker; writes lint-result.json. On FAIL → planner re-dispatches worker.
  4. jg-tester[-fast] — Runs CI + integration; writes test-result.json. On FAIL → planner invokes jg-debugger.
  5. jg-debugger[-high] — Writes debug-diagnosis.json (classification: fix_target | plan_defect | escalate). Planner re-dispatches to worker or subplanner or escalates.
  6. jg-reviewer[-fast|-high] — Only after tester PASS. Writes review-result.json. On FAIL → back to planner.
  7. jg-git — Only after reviewer PASS. Writes git-result.json; optionally archives .pipeline/<issue-id> to .pipeline/completed/<issue-id>.

Subagent Types (Cursor)

When wiring to Cursor subagents, map by role. For Expert tier routing, use the tier-specific variant that matches the classified complexity.

Role Standard Fast (Trivial) High (Complex)
planner jg-planner -- --
subplanner jg-subplanner (skip) jg-subplanner-high
worker jg-worker jg-worker-fast jg-worker-high
tester jg-tester jg-tester-fast jg-tester
reviewer jg-reviewer jg-reviewer-fast jg-reviewer-high
debugger jg-debugger (skip) jg-debugger-high
git jg-git -- --
benchmarker jg-benchmarker (support; invoke on-demand) -- --
linter team-linter (project-specific) -- --