mini-harness
01Plugin Claude Code
A development workflow framework for AI agents, built around three roles and an organiser with four layers of checks.
Plugin Claude Code
A development workflow framework for AI agents, built around three roles and an organiser with four layers of checks.
Agent Claude Code
../../agents/thinker.md.
Agent Claude Code
../../agents/verifier.md.
Agent Claude Code
../../agents/worker.md.
Command Claude Code
A command that reads the Mini-Harness development instructions and then applies them to the Mini-Harness project itself.
Command Claude Code
A command that follows the complete procedure defined in the mh-slideflow skill.
Command Claude Code
A command that reads the complete mh-codeflow skill and follows the workflow it defines.
Hook Claude Code
Runs before the agent uses a tool for Write, Edit and NotebookEdit tool calls, executing role-guard.sh via bash. From zhaodm/mini-harness.
Hook Claude Code
Runs when a session starts, executing session-context.sh via bash. From zhaodm/mini-harness.
Settings file Claude Code
Agent settings declaring 2 hook events (PreToolUse, SessionStart).
Instructions file
Project instructions for Claude Code covering the repository's engineering rules, preparation before coding, preference for simple solutions, and precise edits.
Agent
A central workflow coordinator that manages the overall process, quality checkpoints, and decisions about when human input is needed.
Skill Claude CodeCodex
This skill should be used when in the apply phase, during Worker development, TDD coding, or when building and implementing tasks. Worker development implementation, Verifier audit, batch processing, and SR3 delivery confirmation.
Skill Claude CodeCodex
This skill should be used when the user runs "/mh-run" or when auto-advance is active in code track. Code track full-flow orchestration with clarify→propose→apply→archive pipeline, 3-role spine, state machine, and breakpoint recovery.
Skill Claude CodeCodex
This skill should be used when in the archive phase, during deliverable archiving, lessons learned collection, or knowledge base generation. Product archiving, test case sedimentation, execution metrics, experience collection, and project closure.
Skill Claude CodeCodex
This skill should be used when in the propose phase, during Thinker needs/design/visual phases, at SR1 方案确认, or when designing technical solutions. Thinker requirement analysis, technical design, visual design, plan orchestration, and SR1 review.
Skill Claude CodeCodex
This skill should be used when the user starts a new project requirement, runs "/mh-clarify", or when in the init/clarify phase. Requirement initialization and clarification with track selection, tech stack detection, and test strategy recommendation.
Skill Claude CodeCodex
This skill should be used when Verifier audit fails, during repair cycles, or when fix loops are active. Repair loop orchestration with decideRepair() decision logic, root cause analysis, and repair dispatch.
Skill Claude CodeCodex
This skill should be used when Worker completes coding, before filling code-report, or when running dev-test. Worker self-test SOP with test execution, lint check, build verification, and self-check checklist.
Skill Claude CodeCodex
This skill should be used when the user runs "/mh-ppt" or when working on PPT/slide track. Slide track full-flow orchestration with PPT-specific visual constraints, implementation quality requirements, and verify-ppt.sh validation.
Skill Claude CodeCodex
This skill should be used when running post-verify checks, during Worker delivery verification, or when filling code-report self-check results. Worker delivery pre-check SOP with verify.sh execution, output completeness, unauthorized modification check, and regression check.
Agent
A lead architecture-auditor role for the Mini-Harness project. The auditor independently reviews changes made by the developer and tester using the agreed audit scope and tester evidence.
Agent
A developer role for changing the Mini-Harness framework according to a requirements document. The developer works as an independent sub-agent from the context handed over by a planner.
Agent
An independent tester role for the Mini-Harness project. The tester checks whether a developer’s changes meet the acceptance criteria without relying on the developer’s implementation approach.