Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By TonyWu20
Outcome-Driven Development pipeline for Claude Code — orchestrates planning, implementation with real build feedback, and review through specialized agents and skills. Language-agnostic with per-project configuration.
npx claudepluginhub tonywu20/my-claude-marketplace --plugin odd-pipelineUse this agent when a `/drive-outcomes` orchestrator has produced TASKS.md tasks and you need a specialist to implement those tasks on a git branch with real build feedback. This agent should be invoked for any concrete coding sub-task that requires editing code, running build commands, and fixing errors — the edit→check→fix loop. \n\n<example>\nContext: A drive-outcomes orchestrator has decomposed a phase plan into TASKS.md with task groups.\nuser: \"Implement group-core tasks from TASKS.md\"\nassistant: \"I'll launch the implementation-executor agent to implement these tasks on the branch with build feedback.\" \n<commentary>\nThe drive-outcomes orchestrator delegates a task group to the implementation-executor agent for branch-based implementation with build feedback.\n</commentary>\n</example>\n\n<example>\nContext: A make-judgement review has produced fix-tasks.md with defects to resolve.\nuser: \"Apply the fix directions for the review issues\"\nassistant: \"I'll use the implementation-executor agent to apply the fixes on the branch with the same edit→check→fix loop.\"\n</commentary>\n</example>
Use this agent when you need architectural guidance, code review, or strategic planning for software projects. This includes designing new systems, evaluating existing code structure, refactoring for better design, or when you want a senior engineer's perspective that challenges initial assumptions and applies first-principles thinking.\n\n<example>\nContext: User is starting a new service and wants architectural guidance.\nuser: "I want to build a REST API that handles user authentication and data storage."\nassistant: "Let me use the software-architect agent to design the architecture for this service."\n<commentary>\nThe user needs architectural planning for a new project. The software-architect agent should be invoked to apply first-principles thinking and propose a clean architecture before any code is written.\n</commentary>\n</example>\n\n<example>\nContext: User has written a module and wants a review.\nuser: "Here's my implementation of the repository layer. Can you review it?"\nassistant: "I'll use the software-architect agent to review this code for design quality, separation of concerns, and best practices."\n<commentary>\nA code review request is a prime use case for the software-architect agent, which will evaluate readability, ergonomics, SRP, and test coverage.\n</commentary>\n</example>
Use this agent when code changes, implementations, or fixes need rigorous review against project documentation, architecture, and coding standards. Examples:\n\n<example>\nContext: The user has just implemented a new feature or module in the codebase.\nuser: "I've finished implementing the execution engine in src/engine/executor.py"\nassistant: "Let me launch the strict-code-reviewer agent to verify the implementation against the project architecture and documentation."\n<commentary>\nA significant implementation was completed. Use the Agent tool to launch the strict-code-reviewer to explore the codebase, check docs, and validate the code.\n</commentary>\n</example>\n\n<example>\nContext: The user has applied a bug fix and wants to ensure it's correct and consistent.\nuser: "I fixed the node dependency resolution bug in the workflow engine."\nassistant: "I'll use the strict-code-reviewer agent to verify the fix aligns with the real codebase and doesn't introduce drift from the architecture."\n<commentary>\nA bug fix was applied. Use the Agent tool to launch the strict-code-reviewer to check the fix against actual files and specifications.\n</commentary>\n</example>\n\n<example>\nContext: A subagent or assistant has proposed or applied code changes.\nuser: "The coding agent just refactored the task scheduler."\nassistant: "I'll invoke the strict-code-reviewer agent to double-check the refactor against actual files and prevent any hallucinated changes."\n<commentary>\nAnother agent made changes. Use the strict-code-reviewer to ground-truth the changes against real files and documentation.\n</commentary>\n</example>
Debug an existing fixture-anchored system that passes its acceptance test but produces wrong output. Classifies prior investigation notes, establishes external anchor criteria, applies upstream-audit rule, implements fix with discriminator-value tests, and captures resolution. Use when the user says "/debug-outcomes", "debug this failure", "the test passes but the output is wrong", or describes a symptom in a system that already has fixture files and a passing (but loose) acceptance test.
Interactive skill that helps users define desired outcomes for the next phase through Socratic grilling. Produces a PHASE_PLAN.md with concrete goals, scope boundaries, and success criteria. Uses grill-me + first-principle thinking to question priorities and foundations before committing to a plan. This is the recommended step before `/drive-outcomes`, especially when goals are still vague. Use when the user says "/define-outcomes", "define the outcomes", "clarify what we want", "what should the next phase achieve", "plan the next phase", or wants to decide what the next phase should accomplish before implementing.
Scans a project's existing test suite for placebo test patterns (vacuous assertions, circular round-trip, unbounded thresholds, synthetic-only data). Produces a migration report to help projects adopt the Outcome-Driven Development (ODD) pipeline. Use when migrating from the old TDD pipeline, auditing test quality, or when the user says "/diagnose-tests", "audit my tests", "find placebo tests", or "how healthy are my tests?".
Merged Stage 1+2 — define success criteria, explore against real fixtures, validate, implement, and produce a forensic record. One continuous session with a checkpoint in the middle. Use when the user says "/drive-outcomes <plan-path>", "drive the outcomes for this phase", after /define-outcomes completes, or when a phase plan is already ready for ODD-driven implementation.
File a bug report or feature request for the odd-pipeline itself, with auto-gathered context from the current project and session. Use when the user says "/file-issue", "file a bug", "report an issue", "this is a pipeline problem", or encounters a pipeline defect during daily use.
Uses power tools
Uses Bash, Write, or Edit tools
Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Use this agent when you need expert assistance with React Native development tasks including code analysis, component creation, debugging, performance optimization, or architectural decisions. Examples: <example>Context: User is working on a React Native app and needs help with a navigation issue. user: 'My stack navigator isn't working properly when I try to navigate between screens' assistant: 'Let me use the react-native-dev agent to analyze your navigation setup and provide a solution' <commentary>Since this is a React Native specific issue, use the react-native-dev agent to provide expert guidance on navigation problems.</commentary></example> <example>Context: User wants to create a new component that follows the existing app structure. user: 'I need to create a custom button component that matches our app's design system' assistant: 'I'll use the react-native-dev agent to create a button component that aligns with your existing codebase structure and design patterns' <commentary>The user needs React Native component development that should follow existing patterns, so use the react-native-dev agent.</commentary></example>
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
A comprehensive Fortran development pipeline for Claude Code — orchestrates planning, code review, fix execution, and implementation through specialized agents and skills
Routes Claude Code hook events to a Discord channel — approval requests, notifications, and session stop signals — with interactive Approve/Deny buttons. Also provides slash commands to inspect active sessions and conversation history.
LSP proxy that filters stale diagnostics for Rust, Python, and Fortran
A comprehensive Rust development pipeline for Claude Code — orchestrates planning, code review, fix execution, and implementation through specialized agents and skills
Blocks bare find/grep in Bash calls and detects common rg flag misuse. Enforces fd/rg per CLAUDE.md. Shell file-read and file-edit blocking has moved to the no-shell-file-ops plugin.
Outcome-Driven Development for Claude Code — a language-agnostic pipeline that anchors every implementation decision to external ground truth. Forked from the rust-development-pipeline and generalized to any language with per-project tooling configuration.
In AI-assisted development, TDD creates a closed loop: the agent writes a test based on its understanding, implements to pass it, the test passes, and the implementation is certified as correct — even when it's wrong. The problem: reality never entered the system.
ODD breaks the loop by requiring every assertion to be anchored to something outside the code under test — a fixture file, a reference implementation output, or a published specification.
Goal → Criteria → Tests → Outcomes
| ODD Step | What Changes |
|---|---|
| Define criteria | Expected values must cite a source, not be invented by the agent |
| Explore | Write snippets against real data before committing to criteria |
| Implement | Build to satisfy criteria, not abstract guidance |
| Verify | Run against real fixtures, measure outcomes against criteria |
These patterns look like real tests but provide zero protection:
| Pattern | Why It Fails |
|---|---|
assert!(x.is_finite()) | Passes for any real number — only fails on NaN |
parse(write(x)) == x | Symmetric bugs survive round-trip |
residual < 10.0 (no source) | Bound chosen to pass, not to discriminate |
| Synthetic data matching parser assumptions | Can never fail on format bugs |
ODD flags every one of them. See odd-pattern.md
for the full taxonomy.
Stage 0: /init-project → CONTEXT.md + ADRs (repo constitution)
Stage 0.5 /define-outcomes → PHASE_PLAN.md (goal definition)
Stage 1+2 /drive-outcomes → Session A: forensic TASKS.md
Session B: implementation on branches
Stage 2: /make-judgement → review.md + fix-tasks.md
Stage 3: /debug-outcomes → RESOLUTION.md (debug wrong output)
Aux: /diagnose-tests → test-diagnostics.md (placebo audit)
Aux: /file-issue → GitHub issue (pipeline bug report)
/init-project — Stage 0Settle the repo constitution: domain language, architecture, dependency choices,
coding patterns, tooling configuration. Run once per project before any other
stage. Produces CONTEXT.md (domain glossary + Tooling section) and docs/adr/
(architecture decision records).
/define-outcomes — Stage 0.5Interactive Socratic grilling via first-principle thinking. Questions goals,
scope, assumptions before any code is written. Produces PHASE_PLAN.md.
/drive-outcomes <plan-path> — Stage 1+2The core engine. Session A: grill the user on goals and fixture files, write
exploratory snippets against real data, produce a forensic TASKS.md with
anchored success criteria. Checkpoint pause for user review. Session B:
implement tasks on git branches with real build feedback, auto-review for placebo
patterns before every commit.
/make-judgement <tasks-path> — Stage 2 (Review)Cross-group validation: runtime outcome verification (run tests against declared
fixtures), per-group diff review via strict-code-reviewer, strategic review
via software-architect. Produces review.md, fix-tasks.md, deferred.md.
/debug-outcomes "<symptom>" — Stage 3For systems that pass acceptance tests but produce wrong output. Ten-step process: classify prior claims (EXTERNAL/DERIVED/HYPOTHESIZED), establish anchor criteria, enumerate divergence surface, verify diagnostics via cross-path self-test, apply upstream audit gate, loose-then-tighten cycle, implement fix, capture resolution.
/diagnose-testsStatic scan of existing test suites for placebo patterns. Detects vacuous assertions, circular round-trip, unbounded thresholds, synthetic-only data, and unused fixture files. Produces a migration report — does not modify files.
/file-issueGathers context and files a structured issue on TonyWu20/odd-pipeline via
gh issue create. User reviews before submission.
| Agent | Model | Role |
|---|---|---|
| software-architect | opus | Architectural guidance, code review, first-principles analysis |
| implementation-executor | haiku | Implements tasks on branches with build-feedback loop |
| strict-code-reviewer | sonnet | Rigorous verification against tasks, architecture, and ODD principles |
The pipeline is language-agnostic. All language-specific details live in the
project's CONTEXT.md under a ## Tooling section produced by /init-project:
## Tooling