Team Packs
Team Packs are curated subsets of the 61-role spine. Instead of selecting individual roles, you pick a named pack that already contains the right roles in the right order for a given class of work. Each pack includes mismatch detection: if your task does not fit the pack, it suggests the right one instead.
roleos run --pack=feature "add export command"roleos packs listroleos packs suggest .claude/packets/my-task.mdWhen to use a pack vs. free routing
Use a pack when the work is well-scoped and the domain is clear. Use free routing (roleos route) when the work cuts across domains or you need a custom chain. roleos start picks the right level automatically (mission, pack, or free routing) based on your task description.
The 10 packs
1. Feature Build
Full feature delivery: scope, spec, implement, test, review.
| Roles | Orchestrator, Product Strategist, Spec Writer, Backend Engineer, Test Engineer, Critic Reviewer |
| Optional | UI Designer, Frontend Developer, Security Reviewer |
| Chain | Product Strategist, Spec Writer, Backend Engineer, Test Engineer |
| Orchestrator | Required (multi-role, cross-functional) |
| Artifacts | Scope doc, spec, implementation, test results, verdict |
2. Bugfix / Repair
Diagnose, fix, verify, review. Minimal chain, fast turnaround.
| Roles | Repo Researcher, Backend Engineer, Test Engineer, Critic Reviewer |
| Optional | Frontend Developer, Performance Engineer |
| Chain | Repo Researcher, Backend Engineer, Test Engineer |
| Orchestrator | Not required |
| Artifacts | Repo map / diagnosis, fix implementation, regression tests, verdict |
3. Security Review
Threat model, code review, dependency audit, verdict.
| Roles | Security Reviewer, Dependency Auditor, Critic Reviewer |
| Optional | Backend Engineer, Test Engineer |
| Chain | Security Reviewer, Dependency Auditor |
| Orchestrator | Not required |
| Artifacts | Threat model, code review findings, dependency audit, verdict |
4. Docs / Handbook
Triage raw input, synthesize themes, structure docs, verify metadata, review.
| Roles | Support Triage Lead, Feedback Synthesizer, Docs Architect, Metadata Curator, Critic Reviewer |
| Optional | Repo Translator, Brand Guardian, Release Engineer, Deployment Verifier |
| Chain | Support Triage Lead, Feedback Synthesizer, Docs Architect, Metadata Curator |
| Orchestrator | Not required |
| Artifacts | Classified input, synthesized themes, docs structure, metadata audit, verdict |
5. Launch / Messaging
Plan launch strategy, write copy, review.
| Roles | Launch Strategist, Launch Copywriter, Critic Reviewer |
| Optional | Content Strategist, Community Manager |
| Chain | Launch Strategist, Launch Copywriter |
| Orchestrator | Not required |
| Artifacts | Launch plan, release copy, verdict |
6. Research / Strategy
Frame the decision, gather evidence, synthesize, recommend.
| Roles | Product Strategist, UX Researcher, Competitive Analyst, Feedback Synthesizer, Critic Reviewer |
| Optional | Trend Researcher, User Interview Synthesizer |
| Chain | Product Strategist, UX Researcher, Competitive Analyst, Feedback Synthesizer |
| Orchestrator | Not required |
| Artifacts | Decision frame, friction inventory, competitive landscape, signal synthesis, verdict |
7. Treatment (Repo Polish)
Full repo treatment: research, security, audit, docs, metadata, release, deploy, verify.
| Roles | Repo Researcher, Security Reviewer, Coverage Auditor, Docs Architect, Metadata Curator, Release Engineer, Deployment Verifier, Critic Reviewer |
| Optional | Brand Guardian, Repo Translator, Dependency Auditor |
| Chain | Repo Researcher, Security Reviewer, Coverage Auditor, Docs Architect, Metadata Curator, Release Engineer, Deployment Verifier |
| Orchestrator | Not required (long but sequential, clear handoffs) |
| Artifacts | Repo map, security findings, coverage audit, docs, metadata audit, release, deployment verification, verdict |
8. Deep Audit
Decompose repo into components, audit each deeply, inspect seams from dependency graph, synthesize verdict. Worker count scales with repo manifest via dynamic dispatch.
| Roles | Component Auditor, Test Truth Auditor, Seam Auditor, Audit Synthesizer, Critic Reviewer |
| Optional | Security Reviewer, Dependency Auditor |
| Chain | Component Auditor (×N, parallel) + Test Truth Auditor (×M) → Seam Auditor (×K, from graph) → Audit Synthesizer |
| Orchestrator | Not required |
| Artifacts | Audit manifest, component audit reports, test truth reports, seam audit reports, audit summary, action plan, verdict |
9. Brainstorm
Structured multi-perspective inquiry with traceable disagreement and verdict-bearing output.
| Roles | Context Analyst, User Value Analyst, Mechanics Analyst, Positioning Analyst, Normalizer, Cross-Examiner, Rebuttal Defender, Synthesizer, Judge |
| Chain | 4 Analysts (parallel) → Normalize → Cross-Examine → Rebut → Synthesize → Expand → Judge |
| Orchestrator | Not required |
| Artifacts | Role-native schemas, provenance atoms, cross-exam dispute graph, synthesis, verdict |
Pack reference
| Pack | Key | Roles | Primary use |
|---|---|---|---|
| Feature Build | feature | 6 (+3 optional) | Full feature delivery end-to-end |
| Bugfix / Repair | bugfix | 4 (+2 optional) | Diagnose, fix, test, verify |
| Security Review | security | 3 (+2 optional) | Threat model and dependency audit |
| Docs / Handbook | docs | 5 (+4 optional) | Documentation structure and content |
| Launch / Messaging | launch | 3 (+2 optional) | Launch strategy and release copy |
| Research / Strategy | research | 5 (+2 optional) | Decision framing and evidence gathering |
| Treatment | treatment | 8 (+3 optional) | Full repo polish and publish |
| Deep Audit | deep-audit | 5 (+2 optional) | Manifest-scaled repo audit |
| Brainstorm | brainstorm | 9 | Multi-perspective inquiry with traceable disagreement |
| Dogfood Swarm | swarm | 8 (+0 optional) | Multi-pass convergence with exclusive file ownership |
10. Dogfood Swarm
Multi-pass convergence: three health stages then iterative feature delivery with exclusive file ownership and build gates.
| Roles | Swarm Coordinator, Swarm Backend Agent, Swarm Bridge Agent, Swarm Tests Agent, Swarm Infra Agent, Swarm Frontend Agent, Swarm Synthesizer, Critic Reviewer |
| Optional | (none — domain agents scale from manifest) |
| Chain | Coordinator → [5 domain agents parallel] → Coordinator gate (×4 stages) → Synthesizer → Critic |
| Artifacts | swarm-gate, wave-report (×5 per wave), swarm-final-report, verdict |
| When | You want to move a repo from “works” to “production-ready” through systematic convergence |
| Not for | Single bugs, brainstorms, research, launches, docs-only work |
| Proven | claude-collaborate: 35→129 tests, 106 findings fixed, v1.1.0 shipped |
Mismatch detection
Every pack includes guards that detect when a task does not belong. If you route a bugfix task through the feature pack, the system warns you and suggests bugfix instead. This prevents wasted work from routing through the wrong abstraction.
roleos packs suggest .claude/packets/my-task.md# → Suggested: bugfix (high confidence)# → Mismatch: feature pack is wrong for this task — this is a bug to fix, not a feature to buildCalibration
Packs were calibrated through execution trials and pack comparison trials. Key calibration findings:
- Orchestrator is conditional: only included when the task is multi-role and cross-functional (not for linear chains like bugfix)
- Treatment includes Security Reviewer by default (security before polish)
- Research opens with Product Strategist (framing before investigation)
- Docs has an upstream-synthesis gate (prevents stale documentation)
Outcome calibration records run results and tunes pack weights over time. Use roleos run with different packs to build calibration data for your project.
Combining packs
Packs are not mutually exclusive. A treatment often runs the treatment pack end-to-end, but you can also chain packs manually. The output of each pack is the input for the next. Use roleos review to record the verdict between packs.
Related
- Role Spine — the full 61-role catalog that packs draw from
- Missions — 9 recurring workflows built on top of packs
- Getting Started — run your first task
- Reference — full CLI command reference