How do I install the Harness skill?

Run npx skillmds@latest add d-o-hub/harness in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Harness safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Harness?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Harness free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.

Who published Harness?

d-o-hub (@d-o-hub) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Harness by d-o-hub — Agent Skill

Name: Harness
Author: d-o-hub

Harness

Map the harness-engineering feedforward guides and feedback sensors, and run the self-correction protocol when a computational sensor fires. Use when a sensor fails (do-harness verify), before making code changes, or when setting up agent context for a new task. Triggers: "harness", "sensor fire", "CI failure", "self-correction".

d-o-hub 84c3d27 3 files · 5.0 KB Updated Sep 17, 2026

File contents

Harness Skill

What Is the Harness

Agent = Model + Harness. The harness is the system of feedforward guides (what to do before coding) and feedback sensors (what catches violations after coding).

Feedforward: AGENTS.md and .agents/skills/ — context, constraints, conventions read before coding.
Feedback: do-harness verify — automated checks that fire after code changes. Computational output strictly supersedes LLM self-assessment.

Feedback Sensors

Sensor	Command	Stage
verify	`do-harness verify`	pre-commit (subset) + pre-push + CI (full)
individual	`do-harness verify --only <name>`	targeted re-runs

Sensors are defined in do-harness.toml; run do-harness list to see them.

Self-Correction Protocol

Read the full error output — it includes the failing check.
Classify the failure: formatting / compile / lint / test / config.
Apply the minimal fix — do not refactor unrelated code.
Re-run the specific sensor (do-harness verify --only <name>).
Proceed only when the sensor is green.

Fail-Fast Policy

If the same subtask fails a sensor 3 consecutive times, halt, record the error signature in .do-harness/agent_state.db, and surface a diagnostic.

New-Repo Adoption (Dogfood Rule)

do-harness init (rust pack) scaffolds a minimal crate when no Cargo.toml exists, so init && verify exits 0 on an empty tree; existing crates are never touched, not even with --force.
The generic pack ships zero sensors: its verify pass is vacuous until real [[sensors]] are configured — not evidence.
Never assume the harness works in this repo: re-prove with do-harness verify --format json and read the per-sensor exit codes.

Gotchas

Never trust LLM self-assessment over a computational sensor's exit code.
Fix the sensor that fired; do not refactor unrelated code in the same pass.
An empty sensor suite passes vacuously; that is not evidence.

---name: harnessdescription: Map the harness-engineering feedforward guides and feedback sensors, and run the self-correction protocol when a computational sensor fires. Use when a sensor fails (do-harness verify), before making code changes, or when setting up agent context for a new task. Triggers: "harness", "sensor fire", "CI failure", "self-correction".license: MIT---# Harness Skill## What Is the HarnessAgent = Model + Harness. The harness is the system of feedforward guides (whatto do before coding) and feedback sensors (what catches violations aftercoding).- **Feedforward**: `AGENTS.md` and `.agents/skills/` — context, constraints, conventions read before coding.- **Feedback**: `do-harness verify` — automated checks that fire after code changes. Computational output strictly supersedes LLM self-assessment.## Feedback Sensors| Sensor | Command | Stage ||--------|---------|-------|| verify | `do-harness verify` | pre-commit (subset) + pre-push + CI (full) || individual | `do-harness verify --only <name>` | targeted re-runs |Sensors are defined in `do-harness.toml`; run `do-harness list` to see them.## Self-Correction Protocol1. Read the full error output — it includes the failing check.2. Classify the failure: formatting / compile / lint / test / config.3. Apply the minimal fix — do not refactor unrelated code.4. Re-run the specific sensor (`do-harness verify --only <name>`).5. Proceed only when the sensor is green.## Fail-Fast PolicyIf the same subtask fails a sensor 3 consecutive times, halt, record the errorsignature in `.do-harness/agent_state.db`, and surface a diagnostic.## New-Repo Adoption (Dogfood Rule)- `do-harness init` (rust pack) scaffolds a minimal crate when no `Cargo.toml` exists, so `init && verify` exits 0 on an empty tree; existing crates are never touched, not even with `--force`.- The generic pack ships zero sensors: its `verify` pass is vacuous until real `[[sensors]]` are configured — not evidence.- Never assume the harness works in this repo: re-prove with `do-harness verify --format json` and read the per-sensor exit codes.## Gotchas- Never trust LLM self-assessment over a computational sensor's exit code.- Fix the sensor that fired; do not refactor unrelated code in the same pass.- An empty sensor suite passes vacuously; that is not evidence.

Frequently asked questions