Skill

Pass

Audit score 90

harness-creator

walkinglabs/learn-harness-engineering

Build reliable AI coding agent harnesses with instruction files, state tracking, verification gates, and session handoff.

Source View on skills.sh

What is harness-creator?

Harness Creator helps you make repositories work reliably with coding agents by establishing five core subsystems: instructions (AGENTS.md/CLAUDE.md), state tracking (feature_list.json, progress.md), verification (init.sh), scope boundaries, and session lifecycle management. Use it when agents forget context across sessions, drift out of scope, or claim completion before tests pass.

Create minimal harnesses with bundled scripts (create-harness.mjs) that scaffold instructions, state, and verification files
Audit existing harnesses against five subsystems (instructions, state, verification, scope, lifecycle) and identify bottlenecks
Validate harness structure and generate HTML assessment reports with benchmark scoring
Design multi-session handoff patterns so agents resume reliably with context and progress preserved
Establish verification gates and done criteria to prevent half-finished work and scope creep
Coordinate tool permissions, context budgets, and multi-agent delegation through documented patterns

How to install harness-creator

npx skills add https://github.com/walkinglabs/learn-harness-engineering --skill harness-creator

Prerequisites

Node.js (for bundled scripts: create-harness.mjs, validate-harness.mjs, render-assessment-html.mjs, run-benchmark.mjs)
A local repository to inspect or scaffold
Optional: existing AGENTS.md, CLAUDE.md, feature_list.json, or progress.md files to audit

Claude Code

Cursor

Windsurf

Cline

How to use harness-creator

1.Inspect what already exists in the target repository: instruction files, state files, verification commands, and docs
2.Run `node skills/harness-creator/scripts/create-harness.mjs --target /path/to/project` to scaffold a minimal harness, or `validate-harness.mjs` to audit an existing one
3.Review the generated or scored AGENTS.md, feature_list.json, progress.md, and init.sh files
4.Replace placeholder feature entries with your actual tasks and update done criteria and verification commands
5.For multi-session work, add session-handoff.md and document the end-of-session routine
6.Run verification commands before marking features done and update progress.md with evidence

Use cases

Good for

A coding agent forgets context between sessions and starts each task from scratch—create a session-handoff.md and progress.md to preserve state
An agent claims features are done before tests pass—add explicit verification commands to init.sh and require evidence in feature_list.json
A repository has no clear startup path or working rules for agents—scaffold AGENTS.md or CLAUDE.md with routing, invariants, and definition of done
You need to assess whether a harness is causing agent failures—run validate-harness.mjs to score the five subsystems and pinpoint the weakest area
A multi-agent project needs clear ownership boundaries and coordination—use the multi-agent coordination pattern to delegate features and prevent conflicts

Who it's for

Engineering teams building reliable AI coding agent workflows
Repository maintainers setting up agent-friendly project structure
Teams debugging agent reliability issues across sessions or features
Developers creating reusable agent harnesses as skills or templates

harness-creator FAQ

When should I use Harness Creator vs. just prompting the agent directly?

Use Harness Creator when agents are unreliable across sessions, forget context, drift out of scope, or claim completion prematurely. Direct prompting works for one-off tasks; harnesses make agents reliable for sustained, multi-session work on complex projects.

How minimal can a harness be?

A minimal harness needs AGENTS.md (or CLAUDE.md), feature_list.json, progress.md, and init.sh. Add session-handoff.md, memory patterns, and tool safety only when the user's problem calls for them.

Can I use Harness Creator on an existing project without overwriting files?

Yes. Run validate-harness.mjs first to audit what exists, then use create-harness.mjs with --force only after confirming overwrites are acceptable. The skill prefers inspection and incremental improvement over destructive changes.

What does the benchmark score actually measure?

The benchmark scores the five subsystems (instructions, state, verification, scope, lifecycle) structurally. It proves the bundled scripts work end-to-end but does not measure real agent effectiveness—that requires before/after sessions on representative tasks.

How do I handle multi-agent coordination or tool permissions?

Load the relevant reference files: Multi-Agent Coordination (references/multi-agent-pattern.md) for delegation and parallel work, or Tool Registry & Safety (references/tool-registry-pattern.md) for permissions and concurrency. The skill provides patterns, not automatic setup.

Full instructions (SKILL.md)

Source of truth, from walkinglabs/learn-harness-engineering.

name: harness-creator description: >- Build, audit, and improve harnesses that make AI coding agents reliable: AGENTS.md/CLAUDE.md instruction files, feature/state tracking, verification gates, scope boundaries, session handoff, memory persistence, context budgets, tool-permission safety, and multi-agent coordination. Use this whenever a coding agent is unreliable across sessions — forgets context, drifts out of scope, claims "done" before tests pass, or starts each session inconsistently — or when creating or assessing AGENTS.md, CLAUDE.md, feature_list.json, init.sh, progress.md, or session-handoff files. Reach for it even if the user never says the word "harness." license: MIT

Harness Creator

Use this skill to make a repository easier for coding agents to start, stay in scope, verify work, and resume across sessions. Keep the harness small enough that agents actually follow it.

Not for model selection, prompt tuning in isolation, chat UI design, or general app architecture.

Core Model

Every useful coding-agent harness has five subsystems:

Subsystem	Minimal artifact	Purpose
Instructions	`AGENTS.md` or `CLAUDE.md`	Startup path, working rules, definition of done
State	`feature_list.json`, `progress.md`	Current feature, status, evidence, next step
Verification	`init.sh` or documented commands	Tests/checks the agent must run before claiming done
Scope	Feature dependencies and done criteria	Prevents overreach and half-finished work
Lifecycle	`session-handoff.md`, end-of-session routine	Makes the next session restartable

First Move

Inspect what already exists: instruction files, feature/state files, verification commands, docs, package manifests.
Ask only for missing context that cannot be inferred safely: target agent, desired file name, tolerance for structure, and whether overwriting is allowed.
Prefer a minimal harness first. Add memory, tool safety, multi-agent, or benchmark details only when the user's problem calls for them.

Common Tasks

Create a harness

Use the bundled script when working on a local repository:

node skills/harness-creator/scripts/create-harness.mjs --target /path/to/project

Options:

--agent-file CLAUDE.md for Claude-oriented projects.
--package-manager npm|pnpm|yarn|bun when detection is wrong.
--commands "cmd one,cmd two" for custom verification.
--force only after confirming overwrites are acceptable.

Then explain what was created and how the user should replace placeholder feature entries.

Audit an existing harness

Run:

node skills/harness-creator/scripts/validate-harness.mjs --target /path/to/project

Report the five subsystem scores, the lowest-scoring area, and the first 2-3 changes that would improve reliability. Treat the lowest score as a candidate bottleneck; confirm with failures, logs, or task outcomes before claiming causality.

Produce a report

Use when the user wants a shareable assessment:

node skills/harness-creator/scripts/render-assessment-html.mjs --target /path/to/project
node skills/harness-creator/scripts/run-benchmark.mjs --target /path/to/project --html /path/to/report.html

Be clear that this is a structural benchmark. The benchmark first runs a self-check — it scaffolds a throwaway harness and validates it, proving the bundled scripts work end-to-end — then scores the target and eval coverage. Real effectiveness still needs before/after agent sessions on representative tasks.

When to Read References

Load only the reference needed for the user's problem:

Memory across sessions: Memory Persistence
Reusable workflows as skills: Skill Runtime
Permissions, tools, concurrency: Tool Registry & Safety
Context budget and progressive disclosure: Context Engineering
Delegation and parallel agents: Multi-Agent Coordination
Hooks, startup, long-running work: Lifecycle & Bootstrap
Non-obvious failure modes: Gotchas

Design Rules

Keep the root instruction file short: routing and invariants, not a full manual.
Put project facts in project docs, not in the skill.
Make verification commands explicit and runnable.
Require evidence before marking a feature done.
Use one active feature unless the harness has explicit multi-agent ownership boundaries.
Prefer append/update state files over relying on chat history.
Never hide destructive behavior in scripts; overwrites require explicit user approval.