tools-only Development Harness Plugin - AI-Facing Documentation
Language-agnostic development process harness that orchestrates feature development through a structured 7-stage pipeline. Any language plugin can compose with this harness by providing a language manifest declaring specialist agents and quality gates.
Plugin Identity
Name: development-harness
Version: 0.1.0
Purpose: Provide a reusable, language-independent development workflow based on the Stateless Agent Methodology (SAM) with ARL-derived human touchpoints and Voltron-style language plugin composition.
Design Principles:
- The harness owns the process; language plugins own the specialists
- Every stage produces a file-based artifact (stateless handoff)
- Human escalation follows ARL constraint analysis, not arbitrary checkpoints
- Without a language manifest, the harness falls back to general-purpose agents
How It Works
SAM 7-Stage Pipeline
The harness walks a feature request through seven stages, each producing a named artifact stored in .planning/harness/. Stages gate on artifact completion, not conversation state.
- S1 Discovery - Understand the feature, codebase, and constraints
- S2 Planning + RT-ICA - Generate a plan with information completeness analysis
- S3 Context Integration - Validate the plan against actual codebase state
- S4 Task Decomposition - Break the plan into executable task files
- S5 Execution - Implement tasks using language-appropriate specialists
- S6 Forensic Review - Verify each task against its acceptance criteria
- S7 Final Verification - Certify the feature meets original requirements
The default flow with ARL touchpoint gates is defined in ./skills/development-harness/references/default-development-flow.md.
ARL Human Touchpoints
Not every stage requires human review. The harness uses ARL-derived constraint analysis to decide when to escalate. Escalation triggers include unbound constraints, domain knowledge gaps, high-risk irreversible changes, and novel architecture decisions. Routine changes with existing patterns proceed autonomously.
Details in ./skills/development-harness/references/human-touchpoint-model.md.
Voltron-Style Composition
Language plugins snap into the harness by providing a manifest that maps abstract roles to concrete agents and declares quality gate commands. The harness resolves roles at runtime based on project detection.
Role Resolution
flowchart TD
Start([Feature Request]) --> Detect{Detect Project Language}
Detect -->|pyproject.toml| Python[Python plugin]
Detect -->|package.json| TypeScript[TypeScript plugin]
Detect -->|Cargo.toml| Rust[Rust plugin]
Detect -->|None found| Fallback[General-purpose agents]
Python --> Manifest[Read language manifest]
TypeScript --> Manifest
Rust --> Manifest
Manifest --> Resolve[Resolve roles to agents]
Resolve --> Override{Flow override declared?}
Override -->|Yes| Custom[Use plugin-defined flow]
Override -->|No| Default[Use default SAM pipeline]
Custom --> Configure[Configure orchestrator]
Default --> Configure
Fallback --> Configure
Configure --> Run([Begin S1 Discovery])
The full resolution protocol is documented in ./skills/development-harness/references/role-resolution-protocol.md.
State Management
All artifacts are written to .planning/harness/ using SAM naming conventions.
Token pattern: ARTIFACT:{TYPE}({SCOPE_OR_ID})
File layout example:
.planning/harness/discovery-auth-feature.md- S1 output.planning/harness/plan-auth-feature.md- S2 output.planning/harness/task-001-add-jwt-middleware.md- S4 output per task.planning/harness/execution-001.md- S5 output per task.planning/harness/review-auth-feature.md- S6 output.planning/harness/verification-auth-feature.md- S7 output
This directory coexists with .planning/gsd/ and other planning tools without conflict.
Full conventions in ./skills/development-harness/references/artifact-conventions.md.
Composition Model
flowchart TD
Harness[Development Harness] -->|owns| Process[7-Stage Pipeline]
Harness -->|owns| Touchpoints[ARL Human Touchpoints]
Harness -->|owns| Artifacts[State Management]
LangPlugin[Language Plugin] -->|provides| Manifest[Language Manifest]
Manifest -->|declares| Roles[Specialist Agents]
Manifest -->|declares| Gates[Quality Gate Commands]
Manifest -->|optionally declares| FlowOverride[Custom Process Flow]
Process -->|resolves roles via| Manifest
Process -->|runs gates via| Gates
What the harness owns:
- Process orchestration (stage sequencing, gating, looping)
- Human touchpoint decisions (ARL constraint analysis)
- Artifact management (naming, storage, cross-referencing)
- Fallback behavior (general-purpose agents when no manifest exists)
What language plugins own:
- Specialist agents (architect, test-designer, code-reviewer)
- Quality gate commands (format, lint, typecheck, test)
- Project detection markers (config files, source patterns)
- Optionally, a custom process flow overriding the default pipeline
Language plugin authors should use the template at ./templates/language-manifest-template.md.
The manifest schema is documented in ./skills/development-harness/references/language-manifest-schema.md.
Skills Overview (16)
Main orchestration:
/development-harness- Entry point. Detects language, resolves roles, orchestrates S1-S7.
Workflow stages (7):
/development-harness:discovery- S1 feature and codebase understanding/development-harness:planning- S2 plan generation with RT-ICA/development-harness:context-integration- S3 plan validation against codebase/development-harness:task-decomposition- S4 break plan into executable tasks/development-harness:execution- S5 implement tasks with language specialists/development-harness:forensic-review- S6 verify task completion/development-harness:final-verification- S7 certify feature completion
Planning tools (4):
/development-harness:clear-cove-task-design- Task design methodology/development-harness:generate-task- Generate individual task files/development-harness:planner-rt-ica- Information completeness analysis for planning/development-harness:validation-protocol- Validation patterns and checklists
Implementation:
/development-harness:implementation-manager- Coordinate implementation across tasks
Testing (3):
/development-harness:comprehensive-test-review- Review test coverage and quality/development-harness:analyze-test-failures- Diagnose and categorize test failures/development-harness:test-failure-mindset- Systematic approach to understanding test failures
Agents Overview (11)
Planning and decomposition:
@development-harness:swarm-task-planner- Decompose features into parallel task streams@development-harness:plan-validator- Validate plans for completeness and feasibility
Research and analysis:
@development-harness:feature-researcher- Research feature requirements and prior art@development-harness:codebase-analyzer- Analyze codebase structure and patterns@development-harness:ecosystem-researcher- Research external dependencies and ecosystem
Verification:
@development-harness:feature-verifier- Verify feature meets acceptance criteria@development-harness:integration-checker- Check integration points and compatibility
Context management:
@development-harness:context-gathering- Gather context from codebase and documentation@development-harness:context-refinement- Refine and validate gathered context
Documentation:
@development-harness:doc-drift-auditor- Detect documentation drift from implementation@development-harness:service-docs-maintainer- Generate and maintain service documentation
When to Use
Activate this plugin when:
- Starting feature development in any language project
- Planning an implementation that needs structured decomposition
- Running the full development workflow from discovery through verification
- Working in a multi-language project where process should be consistent
- Needing human touchpoint decisions based on constraint analysis rather than arbitrary gates
Do NOT use when:
- Making a quick fix that does not need staged planning
- Working on documentation-only changes
- The language plugin already provides its own complete workflow (check for flow override in manifest)
References
- Default Development Flow
- Role Resolution Protocol
- Language Manifest Schema
- Human Touchpoint Model
- Artifact Conventions
- Language Manifest Template
Sources
- SAM methodology: https://github.com/bitflight-devops/stateless-agent-methodology
- ARL skill:
plugins/plugin-creator/skills/arl/ - RT-ICA skill:
plugins/python3-development/skills/planner-rt-ica/