Harness Bootstrap Protocol

Harness Bootstrap Protocol

Generate a quorum-governed agent team tailored to the project's domain. Combines Harness's team design patterns with quorum's structural enforcement.

Workflow

Phase 1: Domain Analysis

1. Scan the project: tech stack, data models, key modules, test infrastructure
2. Identify core task types: generation, verification, editing, analysis, migration
3. Check existing quorum configuration: tracks, PRD, CPS, agents, skills
4. Detect language via platform/core/languages/registry.mjs for verify commands

Phase 2: Team Architecture Design

Select an architecture pattern:

Pattern	When to Use
Pipeline	Sequential dependent phases
Fan-out/Fan-in	Parallel independent tasks with result aggregation
Producer-Reviewer	Generation followed by quality review (default for most projects)
Expert Pool	Context-dependent specialist selection
Supervisor	Dynamic task distribution based on progress

Map to quorum's role system:

Project Role	quorum Role	Responsibility
Builder/Developer	implementer	Code generation (headless)
Analyst/Researcher	scout	RTM gap analysis, dependency scanning
Reviewer/QA	self-checker	Mechanical verification (CQ, tests, lint, scope)
Architect/Designer	designer	Design documents, mermaid diagrams
Fixer	fixer	Targeted fixes from audit findings

Ensure consensus coverage: the team must support advocate + devil + judge roles for audit.

Phase 3: Agent Definition Generation

Generate agent files to .claude/agents/{name}.md. Each agent definition must include:

• Role: clear, specific responsibility
• Principles: constraints and guidelines
• Protocol: reference the appropriate agents/knowledge/protocols/*.md
• Error handling: what to do on failure
• Team communication: message targets and data passing

Phase 4: Skill Generation

Generate skills as lightweight manifests referencing agents/knowledge/:

• YAML frontmatter: name, description, knowledge refs, model
• Body ≤ 20 lines — intent + knowledge references only
• Protocol knowledge stays in agents/knowledge/, not in the skill

Phase 5: Integration with Quorum

1. Verify agent files are discoverable by AgentLoader (.claude/agents/ path)
2. Validate consensus requirements: team must cover audit roles
3. If existing quorum track exists, link generated agents to track roles

Phase 6: Validation

1. List generated files and verify structure
2. Check: each agent has a clear role, each skill has frontmatter
3. Verify: no adapter-specific tool names in canonical skills (protocol neutrality)
4. Report: team composition summary with role coverage matrix