SUNCO

Agent Workspace OS — harness engineering for AI coding agents.

85 skills. 7-layer verification. Cross-model review. Zero mistakes.

npx popcoru

What is SUNCO?

In an era where AI agents write code, the builder's job is not writing code — it's setting up the field so agents make fewer mistakes. SUNCO is that field.

• Harness Engineering — lint, health check, and guard enforce quality deterministically (zero LLM cost)
• 85 Skills — every capability is a skill. No hardcoded commands.
• 7-Layer Verification — multi-agent review, guardrails, BDD, permissions, adversarial, cross-model (Codex), human eval
• Skill Absorption — related commands merge via aliases, keeping the surface clean
• Multi-Runtime — Claude Code, Codex CLI, Cursor, Antigravity
• Cross-Model — Layer 6 uses Codex CLI for true cross-family verification (not same-model theater)

Install

npx popcoru

Interactive installer asks: language (EN/KR), runtime (Claude Code/Codex/Cursor/Antigravity/All).

Quick Start

/sunco:new           # idea -> questions -> research -> roadmap
/sunco:discuss 1     # extract decisions for Phase 1
/sunco:plan 1        # create atomic execution plans
/sunco:execute 1     # run plans in parallel waves
/sunco:verify 1      # 7-layer Swiss cheese verification
/sunco:ship 1        # create PR and release

Commands (85)

User Tier (Daily)

| Command | Description | |---------|-------------| | new | Bootstrap from idea to roadmap | | next | Auto-detect and advance to the next step | | do | Natural language -> command routing | | status | Current state, --live for TUI dashboard, --brief for context | | help | Full command guide | | review | Auto-routed review (CEO/eng/design) |

Workflow

| Command | Description | |---------|-------------| | discuss | Extract decisions before planning | | plan | Create execution plans, --assume for approach preview | | execute | Wave-based parallel execution | | verify | 7-layer verification, --coverage for test audit, --generate-tests for AI test gen, --require-codex for strict cross-model | | ship | Create PR after verification | | release | Version bump, changelog, tag, npm publish | | auto | Full autonomous pipeline (discuss -> ship) | | quick | Fast execution with guarantees |

Debug

| Command | Description | |---------|-------------| | debug | Systematic debugging with Iron Law, --parse for diagnostics, --postmortem for forensics | | doc | Document generation (HWPX/markdown), --report for HTML export |

Harness (Zero LLM Cost)

| Command | Description | |---------|-------------| | init | Scaffold .sun/ and .planning/ | | lint | Architecture boundary check | | health | Codebase health score with trends | | guard | Real-time lint-on-change + rule drafting | | agents | Agent instruction analysis |

Session

| Command | Description | |---------|-------------| | note | Idea capture, --todo/--seed/--backlog for task types | | pause / resume | Session management | | where-am-i | Full orientation dashboard | | progress | Phase completion overview |

Architecture

packages/
  core/              — CLI engine, config, state, skill system, agent router
  skills-harness/    — Deterministic backbone (zero LLM cost)
  skills-workflow/   — 35 workflow skills + 9 shared modules
  skills-extension/  — Extension point for user-defined skills
  cli/               — CLI entry point + published artifacts

Published artifacts:
  85 commands          — Skill definitions
  77 workflows         — Process implementations
  23 agents            — Specialized AI agent instructions
  16 references        — Domain knowledge documents
  37 templates         — Artifact templates
  6 hooks              — Lifecycle hooks

v1.4 Highlights — Impeccable Fusion

Shipped 2026-04-20 as [email protected]. See CHANGELOG for full release notes.

Five-milestone track: Frontend Web Fusion (M2), Backend Excellence (M3), Cross-Domain Integration (M4), Rollout Hardening (M5), on top of the M1 Foundation. Zero regression to v0.11.x — every new capability is opt-in via --surface flags.

• /sunco:ui-phase --surface web — vendored Impeccable skill wrapper; /sunco:ui-review --surface web produces IMPECCABLE-AUDIT.md + 6-pillar UI-REVIEW.md
• /sunco:backend-phase --surface {api|data|event|ops} — 4 surface-specific workflows producing API/DATA/EVENT/OPS-SPEC.md with SPEC-BLOCK YAML
• /sunco:backend-review --surface {api|data|event|ops} — 7 deterministic detector rules + LLM review → BACKEND-AUDIT.md
• Cross-domain verify-gate — 4 check types (missing-endpoint HIGH / type-drift HIGH / error-state-mismatch MED / orphan-endpoint LOW) emitted to CROSS-DOMAIN-FINDINGS.md
• /sunco:proceed-gate --allow-low-open — severity × state policy (HIGH hard-block / MED dismissible with ≥50-char rationale / LOW with flag)
• Finding lifecycle (audit_version:2) — state enum expanded to open/resolved/dismissed-with-rationale; HIGH+dismissed structurally rejected
• yaml promoted to direct dependency — runtime SPEC-BLOCK parsing reliability

Docs (this release):

• Impeccable Integration — M2 frontend-web surface
• Backend Excellence — M3 backend track (8 refs + 7 rules + 4 surfaces)
• Cross-Domain Integration — M4 UI ↔ API contract + proceed-gate policy
• Migration v1.4 — non-breaking adoption path from v0.11.x

v0.11.0 Highlights

• Ambient advisor — /sunco:advisor skill + two Claude Code hooks (UserPromptSubmit injection + PostToolUse queue). Deterministic classifier surfaces short Risk: / Suggestion: blocks when risk signals fire. Never writes code, never auto-executes skills.
• First-run model picker — Opus 4.7 / Sonnet 4.6 / Haiku 4.5 / Codex CLI / Custom, with GPT-5 and Gemini 2.5 Pro appearing only when detected. Persists to ~/.sun/config.toml.
• 4-level intervention — silent / notice / guarded / blocker; blocker downgrades to guarded unless blocking=true.
• Queue state machine — pending → surfaced → acknowledged → resolved with 2h TTL and dedupe via SHA1(bucket+file+signals).
• Noise budget — 30-min dedupe per key, 5 visible per session, 1 block per prompt.
• 991 workflow tests, 89/89 contract lint, 85 slash commands, 8 hooks.

v0.10.0 Highlights

• /sunco:orchestrate — dynamic multi-agent router. Deterministic signal-based routing across 8 roles (explorer/librarian/oracle/developer/frontend/docs/verifier/debugger). No fixed pipeline. Clean-room reimplementation of OmO Sisyphus + gstack role discipline; no AGPL code vendored.
• Spec-approval HARD-GATE — /sunco:execute refuses to run without an approved spec. Superpowers brainstorming HARD-GATE now enforced at runtime.
• gstack Sprint Map & OmO Routing Map documented in /sunco:help and pinned by contract-lint.
• 883 workflow tests, 66/66 contract lint, 85 slash commands.

v0.9.0 Highlights

• Superpowers 14-skill parity — every built-in Superpowers skill now has a SUNCO equivalent; brainstorming is vendored verbatim and extended with SUNCO rigor
• Default project-start chain — /sunco:office-hours → /sunco:brainstorming → /sunco:new on all four runtimes
• TDD gate — type: tdd plans get deterministic Red-Green-Refactor enforcement in verify Layer 2
• /sunco:review --fix — auto-routes agreed issues through /sunco:quick and re-verifies
• /sunco:brainstorming --visual — auto-boots the vendored visual companion server
• /sunco:new-skill — scaffolds new SUNCO skills + colocated tests
• 85 slash commands, 847 workflow tests, 59/59 contract lint

v0.8.0 Highlights

• Alias Infrastructure — SkillDefinition.aliases[] with deprecation warnings and default args injection
• Full Absorption — 11 satellite skills merged into 4 absorbers via shared modules (35 skills from 46)
• Codex Layer 6 — CodexCliProvider for true cross-family verification (Claude + OpenAI Codex)
• --require-codex — strict mode flag for pre-ship cross-model gates
• CLI Dashboard TUI — sunco status --live with Ink, 5-section layout, 1Hz polling
• 1,332 tests — 10/10 turbo tasks green

Stats

| Package | Tests | Source Files | |---------|-------|-------------| | @sunco/core | 368 | 244 | | @sunco/skills-harness | 146 | — | | @sunco/skills-workflow | 795 | — | | popcoru (meta) | 23 | — | | Total | 1,332 | 63,466 lines |

Links

• GitHub
• npm

License

MIT