SeongwoongCho/adaptive-harness
A self-improving harness router for Claude Code.
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
Copy this rig
# review before running: this installs third-party code
$ npx degit SeongwoongCho/adaptive-harness/.claude ./rig-adaptive-harness # inspect, then merge into .claude/ MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ npx degit SeongwoongCho/adaptive-harness/skills/harness-registry .claude/skills/harness-registry $ npx degit SeongwoongCho/adaptive-harness/skills/migrate .claude/skills/migrate $ npx degit SeongwoongCho/adaptive-harness/skills/task-taxonomy .claude/skills/task-taxonomy $ npx degit SeongwoongCho/adaptive-harness/skills/using-adaptive-harness .claude/skills/using-adaptive-harness
$ curl -fsSL --create-dirs -o .claude/agents/adversarial-review.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/adversarial-review.md $ curl -fsSL --create-dirs -o .claude/agents/careful-refactor.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/careful-refactor.md $ curl -fsSL --create-dirs -o .claude/agents/code-review.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/code-review.md $ curl -fsSL --create-dirs -o .claude/agents/deep-interview.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/deep-interview.md $ curl -fsSL --create-dirs -o .claude/agents/divide-and-conquer.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/divide-and-conquer.md $ curl -fsSL --create-dirs -o .claude/agents/documentation-writer.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/documentation-writer.md $ curl -fsSL --create-dirs -o .claude/agents/engineering-retro.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/engineering-retro.md $ curl -fsSL --create-dirs -o .claude/agents/evaluator.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/evaluator.md $ curl -fsSL --create-dirs -o .claude/agents/evolution-manager.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/evolution-manager.md $ curl -fsSL --create-dirs -o .claude/agents/migration-safe.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/migration-safe.md $ curl -fsSL --create-dirs -o .claude/agents/parallel-dispatch.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/parallel-dispatch.md $ curl -fsSL --create-dirs -o .claude/agents/performance-optimization.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/performance-optimization.md $ curl -fsSL --create-dirs -o .claude/agents/plan-review.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/plan-review.md $ curl -fsSL --create-dirs -o .claude/agents/pre-landing-review.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/pre-landing-review.md $ curl -fsSL --create-dirs -o .claude/agents/progressive-refinement.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/progressive-refinement.md $ curl -fsSL --create-dirs -o .claude/agents/qa-testing.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/qa-testing.md $ curl -fsSL --create-dirs -o .claude/agents/ralph-loop.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/ralph-loop.md $ curl -fsSL --create-dirs -o .claude/agents/ralplan-consensus.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/ralplan-consensus.md $ curl -fsSL --create-dirs -o .claude/agents/rapid-prototype.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/rapid-prototype.md $ curl -fsSL --create-dirs -o .claude/agents/research-iteration.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/research-iteration.md $ curl -fsSL --create-dirs -o .claude/agents/router.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/router.md $ curl -fsSL --create-dirs -o .claude/agents/security-audit.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/security-audit.md $ curl -fsSL --create-dirs -o .claude/agents/ship-workflow.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/ship-workflow.md $ curl -fsSL --create-dirs -o .claude/agents/simple-executor.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/simple-executor.md $ curl -fsSL --create-dirs -o .claude/agents/spike-then-harden.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/spike-then-harden.md $ curl -fsSL --create-dirs -o .claude/agents/synthesizer.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/synthesizer.md $ curl -fsSL --create-dirs -o .claude/agents/system-design.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/system-design.md $ curl -fsSL --create-dirs -o .claude/agents/systematic-debugging.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/systematic-debugging.md $ curl -fsSL --create-dirs -o .claude/agents/tdd-driven.md https://raw.githubusercontent.com/SeongwoongCho/adaptive-harness/main/agents/tdd-driven.md
This rig commits no guardrails. Here is the community baseline instead — the deny/ask rules most often found across all 7,204 rigs:
{
"permissions": {
"deny": [
"Read(./.env)",
"Read(**/.env)",
"Read(~/.ssh/**)",
"Bash(rm -rf *)",
"Read(**/*.pem)",
"Bash(rm -rf /)",
"Bash(git push --force:*)",
"Bash(sudo *)",
"Read(.env)",
"Bash(rm -rf /*)",
"Read(./.env.*)",
"Read(~/.aws/**)",
"Bash(git push --force*)",
"Bash(rm -rf:*)",
"Read(**/*.key)",
"Read(**/.env.*)",
"Bash(sudo:*)",
"Bash(git reset --hard*)",
"Bash(git reset --hard:*)",
"Read(.env.*)"
],
"ask": [
"Bash(git push:*)",
"Bash(git push *)",
"Bash(git commit:*)",
"Bash(rm *)",
"Bash(rm:*)",
"Bash(git rebase *)",
"Bash(wget *)",
"Bash(npm publish:*)",
"Bash(git commit *)",
"Bash(gh pr merge *)"
]
}
} Skills (4)
Subagents (29)
| adversarial-review model: claude-sonnet-4-6 | Self-attacking implementation executor. Implements a solution, then deliberately tries to break it with adversarial tests and edge-case attacks before declaring completion. |
| careful-refactor model: claude-sonnet-4-6 | Safe refactoring specialist. Applies the Mikado method to restructure code without changing observable behavior. |
| code-review model: claude-opus-4-6 | Multi-perspective code reviewer. Evaluates code across security, quality, performance, and maintainability dimensions. |
| deep-interview model: claude-opus-4-5 | Clarification-first executor. Resolves ambiguous requirements through structured interviews, builds a confirmed spec, then executes against it. |
| divide-and-conquer model: claude-sonnet-4-6 | Decomposition executor. Splits large tasks into independent sub-tasks, solves each in isolation, then integrates and verifies the combined result. |
| documentation-writer model: claude-sonnet-4-6 | Documentation specialist. Reads source truth first, then drafts accurate and well-styled docs, READMEs, API references, and guides. |
| engineering-retro model: claude-sonnet-4-6 | Weekly engineering retrospective specialist. Analyzes commit history, work patterns, and contributor metrics with trend tracking and team-aware praise and growth coaching. |
| evaluator model: claude-opus-4-6 | Scores task results using 6 universal dimensions and collected evidence |
| evolution-manager model: claude-opus-4-6 | Analyzes evaluation history and proposes harness modifications for next-session evolution |
| migration-safe model: claude-sonnet-4-6 | Safe migration specialist. Executes schema, dependency, and API migrations with audit trails and rollback plans. |
| parallel-dispatch model: claude-sonnet-4-6 | Task decomposition specialist. Analyzes a task, identifies independent sub-tasks, and emits a structured decomposition JSON that the orchestrator uses to fan out parallel execution. |
| performance-optimization model: claude-sonnet-4-6 | Performance optimizer. Establishes baselines, profiles bottlenecks, and applies measurement-driven optimizations one at a time. |
| plan-review model: claude-opus-4-6 | Engineering plan reviewer. Challenges scope, reviews architecture/quality/tests/performance one issue at a time, and produces a completion summary with failure mode analysis. |
| pre-landing-review model: claude-sonnet-4-6 | Pre-merge diff reviewer. Analyzes the branch diff against main in two passes — Critical (blocking) and Informational (advisory) — and interactively resolves critical issues. |
| progressive-refinement model: claude-sonnet-4-6 | Iterative quality improvement executor. Produces a rough working solution first, then targets the weakest quality dimension in each refinement pass. |
| qa-testing model: claude-sonnet-4-6 | Systematic QA testing specialist. Tests applications like a real user, computes a health score, and produces a structured report with screenshot evidence and repro steps. |
| ralph-loop model: claude-sonnet-4-6 | Persistent execution loop. Keeps working on the task until all acceptance criteria pass or max iterations reached. |
| ralplan-consensus model: claude-opus-4-6 | Implementation planner with self-review. Analyzes requirements, explores the codebase, creates a structured plan, then challenges its own assumptions. |
| rapid-prototype model: claude-sonnet-4-6 | Fast MVP builder. Implements the minimum viable version of a feature with speed as the primary constraint. |
| research-iteration model: claude-opus-4-6 | Experimental research executor. Runs hypothesis-driven cycles to explore high-uncertainty problems with rigorous measurement. |
| router model: claude-sonnet-4-6 | Classifies tasks via 6-axis taxonomy and selects optimal harness from pool |
| security-audit model: claude-opus-4-5 | Security auditor. Conducts OWASP Top-10 scans, dependency audits, secrets scans, and threat modeling. Produces a prioritized findings report with fix recommendations. |
| ship-workflow model: claude-sonnet-4-6 | Automated ship/release specialist. Merges main, runs tests, reviews the diff, bumps version, generates changelog, creates bisectable commits, pushes, and opens a PR. |
| simple-executor model: claude-sonnet-4-6 | Lightweight executor for trivial, well-defined local tasks. No planning overhead — read, change, sanity-check, done. |
| spike-then-harden model: claude-sonnet-4-6 | Two-phase executor. Builds a fast throwaway prototype to learn the problem space, then rewrites for production quality using spike learnings. |
| synthesizer model: claude-opus-4-6 | Merges independent codebases from ensemble worktrees into a single working project via file-level comparison, cherry-pick, and verified integration. |
| system-design model: claude-sonnet-4-6 | Multi-component system architect and implementer. Designs service decomposition, infrastructure, and async patterns before building each component with integration verification. |
| systematic-debugging model: claude-sonnet-4-6 | Root cause analysis executor. Diagnoses bugs through structured reproduce-isolate-fix-verify phases, never guessing. |
| tdd-driven model: claude-sonnet-4-6 | TDD workflow executor. Implements changes using strict red-green-refactor cycles with enforced test coverage gates. |
Slash commands (4)
/eval/evolve/run/status
Similar rigs
kenneth-liao/claude-agent-sdk-intro
—
Pragmatist 73 tok ·
DawnEver/claude-code-config
—
Pragmatist 6.7k tok ·
guhcostan/claude-mega-brain
OKF-powered knowledge context for Claude Code — injects your project's knowledge base at every session
Minimalist 339 tok ·
tasksmd/tasks.md
A lightweight spec for AI agent task queues — the companion to AGENTS.md
Pragmatist 4.0k tok ·