Skip to content

Medir y mejorar

Todo en esta guía condensa en un comando:

bash
npx harness-score

El escáner recorre tu repositorio (solo filesystem — sin LLM, sin red, sin telemetría), corre 36 checks deterministas en cualquier herramienta de IA y reporta un nivel de madurez con los gaps exactos al siguiente:

  harness-score v1.0.0  /work/my-app

  Maturity: L2 · Guided   Score: 66/108 (61%)
  Detected: Cursor, Claude Code

  Context & Guides     ████████████████░░░░  80%
  Skills & Commands    █████████████░░░░░░░  67%
  Hooks & Guardrails   ░░░░░░░░░░░░░░░░░░░░   0%
  ...

  To reach L3: sensors ≥ 60%; ci ≥ 50%

Multi-herramienta: El escáner reconoce artefactos de harness de Cursor, Claude Code, Windsurf, Cline, Continue y otras herramientas vía semántica OR — si configuras cualquiera, Harness Score lo cuenta. Más en Soporte multi-harness.

Instalación

bash
npx harness-score                                       # no install
npm install -g harness-score                            # global binary
npm install --save-dev harness-score                    # pinned devDependency

También espejado en GitHub Packages (@paladini/harness-score) y JSR para proyectos Deno/Bun.

Usándolo como biblioteca

La CLI es un wrapper fino sobre API programática totalmente tipada — útil para dashboard custom, bot o herramienta que quiere el Report crudo en lugar de parsear salida de terminal:

ts
import { score } from 'harness-score';

const report = score('/path/to/repo');
console.log(report.level.name, report.score.percent, report.dimensions);
// With global scopes: score(path, { scopeFlags: ['user'] })
console.log(report.effective.level.index);

Report, Check, CheckResult, DimensionScore, LevelInfo, ScoreSnapshot y toda forma shipan como declaraciones TypeScript — resueltas vía campo "types" explícito, para editores y tsc sin config extra. Bloques de nivel inferior también se exportan, para lo que score() no cubre directamente:

ts
import { score, computeDiff, renderMarkdown } from 'harness-score';

const report = score('/path/to/repo');
const markdown = renderMarkdown(report);          // same renderer the CLI's --md uses

Configuración del scan

Por defecto el escáner mide solo la madurez del repositorio — el harness que viaja con el código y se reproduce en CI. Opcionalmente incluye árboles de harness a nivel usuario o compartidos para un puntaje effective (lo que el agente probablemente ve en el laptop de un desarrollador).

json
{
  "scopes": { "user": false, "system": false },
  "extraRoots": [{ "id": "team-shared", "path": "../shared-harness" }],
  "gate": "maturity"
}

Guárdalo como .harness-score.json en la raíz del scan, o pasa --config <file>. Referencia completa: Métricas y códigos — configuración.

bash
harness-score --scope user              # repo + ~/.cursor, ~/.claude, …
harness-score --scope user,system
harness-score --gate effective --min-level 2   # gate on effective score

El reporte de terminal muestra Maturity (repo) y Effective (cuando difieren). CI debe mantener gate: maturity salvo que corras intencionalmente en runners self-hosted con harness de usuario poblado.

Referencia CLI

bash
harness-score [path]              # human report (default: current directory)
harness-score --json              # full report as JSON
harness-score --md report.md      # markdown report (use "-" for stdout)
harness-score --badge badge.svg   # SVG pill: harness + detected level (L0–L4)
harness-score --min-level 3       # exit 1 if below L3 — the CI gate (uses gate mode)
harness-score --diff base.json    # compare maturity against a previous --json report
harness-score --config .harness-score.json
harness-score --scope user        # include user-level harness in effective score
harness-score --gate maturity     # or effective — which score --min-level uses

Seguir puntuación en el tiempo

--diff <file> compara el scan actual contra reporte baseline guardado de ejecución --json anterior — deltas de nivel y puntuación, movimiento por dimensión y exactamente qué checks cambiaron:

bash
harness-score --json > baseline.json   # save today's report
# ...later, after changes...
harness-score --diff baseline.json     # see what moved
  Compared to baseline:
    Level: L2 · Guided → L3 · Sensing (+1)
    Score: 61/108 (56%) → 84/108 (78%) (+22pp)
    Sensors & Feedback   20% → 90% (+70pp)
    Newly passing: SNS-01, SNS-02, SNS-04, CI-01, CI-02

--diff funciona con --json (agrega current/baseline/diff al payload) y --md (agrega sección "Compared to baseline") — es lo que la GitHub Action usa para comentar "puntuación subió de L2 a L3" en PRs.

El plugin Cursor

Instala Harness Score desde su directorio de plugin en este repo (listado Cursor Marketplace enviado y pendiente de revisión — este enlace moverá allí cuando esté live) y obtienes:

  • /harness-audit — corre el escáner en el workspace abierto y hace que el agente presente el reporte con las principales remediaciones.
  • La skill harness-engineering — cuando dices "arréglalo" o "mejora mi harness", el agente sabe escribir los artefactos faltantes siguiendo las recetas de esta guía.

El análisis en sí es siempre la CLI determinista; el modelo solo presenta resultados y aplica correcciones que pidas.

El gate de CI

Los harnesses regresan en silencio — alguien borra hooks.json en una limpieza, una rule se pudre. Traba tu nivel en CI:

yaml
- name: Harness gate
  run: npx -y harness-score --min-level 3

O usa la action empaquetada, que también emite el badge:

yaml
- uses: paladini/harness-score@main
  with:
    min-level: '3'
    badge: 'harness-badge.svg'

Muestra tu madurez

Harness Score entrega dos formatos SVG con marca en el mismo lenguaje visual de las barras de progreso del escáner — sin shields.io, sin servicio pago, sin red al renderizar:

FormatoArchivosMuestraMejor para
Badgeharness-badge.svg o badge-l0.svgbadge-l4.svgharness · L4Fila README (píldora 112×20)
Share cardcard-l0.svgcard-l4.svgBanner completo con nombre de nivelPosts sociales, hero del repo (860×240)

El badge muestra solo el nivel (L0L4). Nombres de nivel (Unharnessed, Guided, …) viven en share cards y en la salida del escáner.

La píldora se ve idéntica si CI regenera o fijas archivo estático — solo cambia el cableado.

Badge — auto-actualizando (recomendado)

harness-score --badge escribe SVG para el nivel que detecta el escáner. Configúralo en CI una vez; la imagen en README se actualiza conforme mejora el harness.

yaml
# .github/workflows/harness.yml
name: Harness Score
on: { push: { branches: [main] } }
permissions: { contents: write }
jobs:
  harness:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: paladini/harness-score@main
        with: { badge: 'harness-badge.svg' }
      - uses: JamesIves/github-pages-deploy-action@v4
        with: { branch: badges, folder: ., clean: false }

Referencia el archivo publicado desde tu README — recetas copy-paste completas en Snippets de embed:

md
<img alt="Harness Score" src="https://raw.githubusercontent.com/<you>/<repo>/badges/harness-badge.svg" height="20">

Set height="20" on the <img> so the pill lines up with npm/CI shields in the same row (112×20 SVG; level only — score percent stays in the CLI report).

Dogfood example (this guide's live badge on GitHub Pages):

Badge (this repo)

Harness Score

The matching share card for the detected level is published as harness-card.svg (currently L4 for this repository):

Harness Score L4 · Self-correcting

Badge — pin a level

Same pill, static file — pick badge-l0.svgbadge-l4.svg if you do not want CI to regenerate the image. See Embed snippets for Markdown, HTML, iframe, JSX, and more.

Share card

For a hero image or social post, use the banner card — it includes the level name (Unharnessed, Guided, …):

LevelBadgeShare card
L0 · Unharnessedbadge-l0.svgcard-l0.svg
L1 · Documentedbadge-l1.svgcard-l1.svg
L2 · Guidedbadge-l2.svgcard-l2.svg
L3 · Sensingbadge-l3.svgcard-l3.svg
L4 · Self-correctingbadge-l4.svgcard-l4.svg

All badge levels (112×20)

L0L1L2L3L4

Share card example (860×240)

L4 · Self-correcting

Descarga cualquier nivel de la tabla — las cards incluyen el nombre del nivel.

Snippets de embed

Recetas copy-paste para compartir. Reemplaza placeholders:

PlaceholderBadge auto-actualizandoBadge fijo (nivel {N})Share card
{BADGE_URL}https://raw.githubusercontent.com/{owner}/{repo}/badges/harness-badge.svghttps://paladini.github.io/harness-score/maturity/badge-l{N}.svg
{CARD_URL}https://paladini.github.io/harness-score/maturity/card-l{N}.svg
{LINK}Tu repo o https://paladini.github.io/harness-score/IgualIgual

{N} es 04. Badge live de este repo (sin CI en tu fork): https://raw.githubusercontent.com/paladini/harness-score/main/docs/public/harness-badge.svg

Tamaño del badge: 112×20 — siempre define height="20" (o height={20}) para que la píldora alinee con badges shields.io en la misma fila.

Badge — Markdown

Image only (GitHub, GitLab, dev.to — use HTML if plain ![]() stretches):

md
<img alt="Harness Score L4" src="{BADGE_URL}" height="20">

Linked (clickable):

md
[![Harness Score L4]({BADGE_URL})]({LINK})

Reference-style:

md
[![Harness Score][hs-badge]][hs-link]

[hs-badge]: {BADGE_URL}
[hs-link]: {LINK}

Badge — HTML

html
<img alt="Harness Score L4" src="{BADGE_URL}" height="20" width="112">

Linked:

html
<a href="{LINK}">
  <img alt="Harness Score L4" src="{BADGE_URL}" height="20" width="112">
</a>

Badge — iframe

For CMS or wikis that only allow iframes (not <img>):

html
<iframe
  src="{BADGE_URL}"
  title="Harness Score L4"
  width="112"
  height="20"
  style="border:0;overflow:hidden"
></iframe>

Badge — SVG object / embed

html
<object data="{BADGE_URL}" type="image/svg+xml" width="112" height="20">
  <a href="{BADGE_URL}">Harness Score L4</a>
</object>
html
<embed src="{BADGE_URL}" type="image/svg+xml" width="112" height="20" />

Badge — JSX / React

jsx
<a href="{LINK}">
  <img
    alt="Harness Score L4"
    src="{BADGE_URL}"
    height={20}
    width={112}
    style={{ verticalAlign: 'middle' }}
  />
</a>

Badge — AsciiDoc

asciidoc
image:{BADGE_URL}[Harness Score L4,link={LINK},height=20]

Badge — BBCode (forums)

text
[url={LINK}][img]{BADGE_URL}[/img][/url]

Badge — direct URL

Paste in chat, Notion image block, Slack, Discord, or any tool that accepts a raw image URL:

text
{BADGE_URL}

Share card — Markdown / HTML

Banner for README hero, blog posts, or social previews ({N} = 04):

md
[![Harness Score L4 · Self-correcting]({CARD_URL})]({LINK})
html
<a href="{LINK}">
  <img
    alt="Harness Score L4 · Self-correcting"
    src="{CARD_URL}"
    width="560"
    style="max-width:100%;height:auto;border-radius:8px"
  />
</a>

Share card — iframe

html
<iframe
  src="{CARD_URL}"
  title="Harness Score L4 · Self-correcting"
  width="560"
  height="157"
  style="border:0;max-width:100%"
></iframe>

Share card — direct URL

text
{CARD_URL}

Worked example (pinned L3 badge)

md
<a href="https://paladini.github.io/harness-score/">
  <img alt="Harness Score L3" src="https://paladini.github.io/harness-score/maturity/badge-l3.svg" height="20">
</a>
html
<iframe
  src="https://paladini.github.io/harness-score/maturity/badge-l3.svg"
  title="Harness Score L3"
  width="112"
  height="20"
  style="border:0"
></iframe>

shields.io fan? Your Action can also write a small JSON file and point a shields endpoint at it ({ "schemaVersion": 1, "label": "harness", "message": "L3", "color": "brightgreen" }). The brand SVGs above are self-contained and need no third party.

Catálogo de checks

Cada check que corre el escáner, con receta de remediación. Los IDs de check son estables; la CLI enlaza cada fallo a su entrada aquí.

Context & Guides (20 pt)

CTX-01 · Agent context file present — 4 pts

An AGENTS.md (or CLAUDE.md / GEMINI.md) exists at the repository root. Corrección: create AGENTS.md answering: what is this project, how do I build and test it, what conventions hold, what must I never touch. Recipe in capítulo 3.

CTX-02 · Context file is substantive — 3 pts

≥20 meaningful lines and ≥2 headings — a stub scores nothing. Corrección: cover layout, build & test commands, conventions, and no-go zones. Commands over descriptions; point to rules instead of pasting them.

CTX-03 · Scoped rules in use — 4 pts

At least one scoped rule file for any supported tool (e.g. .cursor/rules/*.mdc, .windsurf/rules/*.md, .clinerules/*.md, .continue/rules/*.md, .github/instructions/*.instructions.md, .agents/rules/*). Nested context files in subdirectories (AGENTS.md, CLAUDE.md, GEMINI.md anywhere below the root) also count — they are directory-scoped rules in tools like Claude Code and Codex. Corrección: start with one short always-on rule holding your non-negotiables, then add path-scoped rules per area (or nested context files per subtree).

CTX-04 · Rules have valid frontmatter — 3 pts

Every rule declares activation metadata (description, globs/trigger/paths/applyTo, or alwaysApply). Rules a tool auto-loads without metadata — .continue/rules/* and nested context files — pass by construction. Corrección: add the frontmatter block; without it the agent can't decide when the rule applies.

CTX-05 · Rules are scoped — 2 pts

Not every rule is blanket always-on. Nested context files count as scoped — they apply only to their subtree. Corrección: scope rules to paths (globs:, trigger: glob, paths:, applyTo:) so they load only when relevant — every always-on rule taxes every request's context.

CTX-06 · No bloated rules — 2 pts

No single rule exceeds 500 lines. Corrección: split by concern, or move procedural content into a skill.

CTX-07 · README present — 1 pt

Corrección: add a README.md; it's the first orientation document for humans and a fallback for agents.

CTX-08 · No legacy .cursorrules — 1 pt

The deprecated single-file format is absent (or modern scoped rules also exist). Corrección: migrate .cursorrules content into scoped rules for your tool.

Skills & Commands (17 pts)

SKL-01 · At least one skill — 4 pts

A SKILL.md under .cursor/skills/<name>/, .claude/skills/<name>/, or .agents/skills/<name>/. Corrección: package your most repeated procedure (deploy, release, migration) as a skill — capítulo 3.

SKL-02 · Skills declare name and description — 3 pts

Frontmatter with name: and description: on every skill. Corrección: the agent decides whether to load a skill from these two fields alone; without them the skill is invisible.

SKL-03 · Explicit workflows/commands defined — 3 pts

Command or workflow files (.cursor/commands/, .windsurf/workflows/, .claude/commands/, .continue/prompts/, .zed/commands/, .agents/workflows/). Corrección: encode workflows you trigger deliberately (/review, /release) as command/workflow files.

SKL-04 · Skill descriptions are trigger-worthy — 2 pts

Descriptions ≥40 characters. Corrección: write descriptions as trigger conditions — "Use when the user asks to deploy or release; covers tagging, pipeline, rollback, smoke tests."

AGT-01 · Custom subagent defined — 3 pts

A subagent file under .cursor/agents/, .claude/agents/, or .opencode/agents/. Corrección: package a purpose-built subagent for a job the primary agent should delegate (planning, review, release) — see Subagents en el capítulo 2.

AGT-02 · Subagents declare name and description — 2 pts

Frontmatter with name: and description: on every subagent definition. Corrección: the parent agent decides whether to delegate from these two fields alone; without them the subagent is never invoked.

Hooks & Guardrails (14 pts)

HKS-01 · Hooks configuration present and valid — 4 pts

.cursor/hooks.json or .claude/settings.json (hooks key) exists and parses as JSON. Corrección: create hooks config and grow from the recetas en el capítulo 5.

HKS-02 · Known events, version declared — 2 pts

Version/metadata present; every registered event is documented for your tool (Cursor lifecycle events, or Claude Code PreToolUse/PostToolUse). Corrección: typo'd event names fail silently — check against the event list in capítulo 2.

HKS-03 · Gate hook guards risky operations — 4 pts

A gate hook registered (Cursor: beforeShellExecution, beforeMCPExecution, preToolUse, or beforeReadFile; Claude Code: PreToolUse). Corrección: gate de deny de comandos destructivos del capítulo 5 — rules en prosa son pedidos; gates son hechos.

HKS-04 · Feedback hook observes output — 2 pts

A feedback hook registered (Cursor: afterFileEdit, postToolUse, …; Claude Code: PostToolUse). Corrección: format-and-lint on edit gives the agent instant feedback inside the session.

HKS-05 · Hook scripts committed — 2 pts

Scripts referenced by the hooks config exist in the repository. Corrección: commit them; a hook pointing at a missing script fails open on every machine but the author's.

Sensors & Feedback (20 pts)

SNS-01 · Test runner configured — 6 pts

A real test script/config (vitest, jest, pytest, go test, cargo test…). Corrección: wire up the runner with one obvious entry point and document it in AGENTS.md — tests are how the agent verifies its own work.

SNS-02 · Linter configured — 5 pts

eslint/biome, ruff, golangci-lint, rubocop, or equivalent. Corrección: every convention expressible as a lint rule stops needing prose.

SNS-03 · Type checking in place — 4 pts

tsconfig (ideally strict: true), mypy/pyright, or a statically typed language. Corrección: the type checker is the only sensor that reviews every agent edit for free — capítulo 4.

SNS-04 · Formatter configured — 3 pts

prettier/biome, black/ruff-format, gofmt/rustfmt. Corrección: formatting noise in diffs hides real mistakes from review.

SNS-05 · Test files exist — 2 pts

At least one actual test file in the tree. Corrección: a configured runner with zero tests is a green light nobody earned.

CI Feedback (14 pts)

CI-01 · CI pipeline configured — 4 pts

GitHub Actions workflow (or GitLab/CircleCI/Jenkins equivalent). Corrección: add .github/workflows/ci.yml running your sensors on every push.

CI-02 · CI runs the tests — 4 pts

Corrección: no agent-authored change should be mergeable without the suite firing.

CI-03 · CI runs lint/typecheck — 3 pts

Corrección: cheap computational sensors belong on every push — keep quality left.

CI-04 · Pre-commit checks installed — 3 pts

husky/lint-staged, pre-commit, or lefthook. Corrección: the earliest feedback a commit can get; catches what on-edit hooks missed before it enters history.

Hygiene & Safety (23 pts)

HYG-01 · .gitignore present — 2 pts

Corrección: agents commit what they see; make build output and local state invisible.

HYG-02 · .gitignore covers env files — 3 pts

A .env pattern in .gitignore. Corrección: add .env and .env.* (allow !.env.example) — the cheapest guardrail in existence.

HYG-03 · No unprotected .env files — 4 pts

No real env files in the tree unless gitignored (templates are fine). Corrección: move secrets out; keep .env.example documenting required variables.

HYG-04 · MCP config free of credentials — 4 pts

No credential signatures in MCP config (.cursor/mcp.json, .mcp.json, .agents/mcp_config.json). Corrección: use ${ENV_VAR} interpolation — an inlined key in MCP config is a secret published to every clone.

HYG-05 · License present — 2 pts

Corrección: add a LICENSE; required for open-source use and plugin marketplaces.

HYG-06 · No secrets in harness files — 2 pts

AGENTS.md, rules, and hooks config are clean of token signatures. Corrección: these files are loaded into model context every session — a key there is exfiltrated by design.

HYG-07 · Lockfile committed — 3 pts

package-lock.json, uv.lock, Cargo.lock, go.sum, or equivalent. Corrección: reproducible installs mean your sensors test the same dependency tree everywhere.

HYG-08 · MCP config uses env interpolation for credentials — 3 pts

An MCP config file is valid, and any credential-shaped field (token, key, secret, password…) uses ${ENV_VAR} interpolation rather than a literal. The positive complement to HYG-04 — a repo with no MCP setup earns nothing here, same as any other bonus check. Corrección: reference secrets as "${VAR_NAME}" and document required variables in .env.example.

Plan de mejora práctico

Partiendo de un repo de producto típico en L0, una sesión enfocada por nivel:

  1. → L1 (una tarde). Escribe AGENTS.md (CTX-01/02). Incluye comandos build/test aunque los sensores sean débiles — el agente los usará.
  2. → L2 (un día). Tres rules con alcance + una skill para el procedimiento más repetido (CTX-03…06, SKL-01/02). Corrige higiene: gitignore, env files, licencia (HYG-01…05).
  3. → L3 (el trabajo real, si faltan sensores). Test runner + linter + tipos strict + ci.yml corriendo los tres (SNS-*, CI-01…03). Si ya los tienes, este nivel es gratis.
  4. → L4 (una mañana). Los dos hooks del capítulo 5 — un gate, un formatter — commiteados con scripts (HKS-*), pre-commit (CI-04), luego --min-level 4 en CI para que nunca regrese.