From agent-swarming
Multi-agent collaboration patterns + agent roles + swarming protocols + cross-model verification for Orryx Group development. Triggers on multi-agent task design, "spawn agents", "parallelise this work", or planning a complex delivery. Canonical Orryx pattern; lifted from Triora.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-swarming:agent-swarmingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Cross-cutting pattern for Orryx Group development. Originally Triora's; promoted to cross-cutting in Wave 3 of the architecture migration.
Cross-cutting pattern for Orryx Group development. Originally Triora's; promoted to cross-cutting in Wave 3 of the architecture migration.
Supports multi-agent swarming between Claude and ChatGPT for autonomous development.
| Agent | Model | Primary Role | Tier source |
|---|---|---|---|
| Engineer | Claude | Code implementation, testing | sonnet |
| Architect | Claude | Design, architecture decisions | opus |
| Tester | Claude | Test generation, coverage | sonnet |
| Researcher | Claude Opus | Deep research, knowledge gathering | opus |
| Designer | Claude | UI/UX design, accessibility | sonnet |
| Hygiene | Claude | Repo cleanliness, archival, consolidation | sonnet |
| Reviewer | GPT-4 | Code review, verification | n/a (external) |
| QA | GPT-4 | Integration testing, acceptance | n/a (external) |
| Explorer | Claude | Read-only investigation | haiku, permissionMode: plan |
| Doc-writer | Claude | Documentation generation | haiku |
Tier assignments are canonical via foundation/model-tiering-convention. Configs typically live in agents/claude/<role>.yaml and agents/chatgpt/<role>.yaml per repo.
The Researcher Agent runs independently in parallel to:
research/outputs/Launch: select a prompt, refine, start a new Claude Code session with /model opus, execute research while main agent continues.
For critical outputs (especially regex / async / security / matching logic):
Claude (generate) → GPT-4 (verify) → Compare
Agreement → High confidence, proceed
Minor disagreement → Iterate and refine
Major disagreement → Escalate to human
See foundation/cross-model-verification for the formal protocol.
## Agent Handoff: [Task ID]
### Context
- Task: [Description]
- Phase: [Current phase]
- From: [Agent ID] / To: [Agent ID]
### Completed Work
- [x] What was done / Files changed
### Verification Requested
- [ ] Correctness / Security / Tests
### Confidence: [0.0-1.0]
| Pattern | Description | Use Case |
|---|---|---|
| Sequential | A → B → C | Standard feature development |
| Parallel | [A + B] → Merge | Independent components |
| Verification | A → B (verify) | Critical decisions |
| Debate | A + B → Arbitrate | Architecture choices |
| Refinement | A → B (review) → A (fix) → Loop | Complex logic |
Parallel agents MUST work on separate branches. This is non-negotiable.
Concurrent agents on the same branch cause commit collisions. This was learned the hard way in Triora when a compliance agent stepped on the engineer's commits, resulting in lost work and a messy force-push recovery.
git branch and ensure no other agent targets your branchfeat/agent-name-task or fix/agent-name-issuegit status and git log --oneline -3 before committingcurrent_branch=$(git branch --show-current)
echo "Committing to: $current_branch"
git status
git stashgit pull --rebasegit stash poporchestration/cpd-protocol — Continuous Parallel Development (5-stream pattern)orchestration/ralph-loop — autonomous-loop patternorchestration/git-worktrees — workspace isolation patternfoundation/model-tiering-convention — which tier for which rolefoundation/cross-model-verification — the verifier handoff schemaD:\Clinical.Trials\.claude\skills\agent-swarming\SKILL.mdProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub alexmclaren/orryx-knowledge --plugin agent-swarming