agentsclimarketplace

Harness build

Skill atlax-tech/harness-armor/skills/harness-build

Self-improving, tool-first Harness Engineering Agent Skills for Claude Code, Codex, Cursor and TRAE. Build, audit, update and verify AGENTS.md-based repositories.

Install
npx -y skills add atlax-tech/harness-armor --skill harness-build

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Read substantive product, requirements, design, and planning sources in a docs-only repository and build a project-specific, traceable Harness without business code. Use when the repository has product documents but no implementation, or when harness-build is invoked. Do not use for empty repositories, legacy code promotion, ordinary build commands, or managed-Harness updates.

The file declares its own license as CC-BY-NC-4.0. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.9 KB, as published. Nobody here has run it

Build a product-specific Harness from documents

Turn real product sources into a usable engineering knowledge system while keeping every conclusion traceable.

Inputs

  • Repository root containing substantive product sources and no business code.
  • Optional user priorities or source precedence. Record these as explicit user evidence; do not silently choose a preferred document.

Load the contract

  1. Treat this file's directory as SKILL_ROOT.
  2. Read references/evidence-extraction.md.
  3. Resolve shared resources at SKILL_ROOT/../../shared or SKILL_ROOT/../.harness-armor.
  4. Read spec/harness-engineering-v1.md, schemas/, and templates/ from that root. Stop if the installation is incomplete.

Applicability gate

  1. Read applicable AGENTS.md files.
  2. Run python scripts/scan_repository.py <root> and inspect all likely product sources.
  3. Confirm substantive product documentation exists and no actual business code exists. Generated snippets inside documents do not by themselves count as an implementation.
  4. Route EMPTY to harness-init, code to harness-promotion, managed state to harness-update or harness-check, and conflicting repository structure to harness-check.

Evidence extraction

  1. Inventory and classify product, requirements, design, research, roadmap, acceptance, and constraint sources.
  2. Read every relevant source, not only README.
  3. Build the ledger in assets/fact-ledger.md for: product goals, users, scenarios, functional requirements, business rules, design requirements, technical constraints, non-functional requirements, acceptance conditions, and plan steps.
  4. Mark each item CONFIRMED, INFERRED, UNRESOLVED, or CONFLICTED. Include source path and heading/locator for confirmed and conflicted items. Explain inference reasoning.
  5. Preserve both sides of every conflict. Do not demote a required feature into a roadmap item or simplify a user journey to make implementation easier.
  6. Run python scripts/fingerprint_sources.py <root> <paths...> for the final source set.

Harness plan and authorization

  1. Propose the file-level Harness plan: short AGENTS.md, only the necessary focused docs, .harness state, and decision/log maps.
  2. An explicit build request authorizes creation of missing Harness files in the plan. It does not authorize overwriting any existing file. Present collisions and request approval for a managed section or alternate path.
  3. Keep source documents user-owned. Never replace them with generated summaries.

Write workflow

  1. Write current product requirements without adding facts.
  2. Write architecture and design documents that separate required constraints, evidence-backed inferences, open decisions, and recommendations.
  3. Write development, testing, acceptance, and roadmap documents only to the level supported by evidence. Unknown commands stay unresolved.
  4. Write a concise AGENTS.md knowledge map with source-of-truth pointers and verified commands only.
  5. Write .harness/manifest.json, source-index.json, and unresolved.json. Record source digests and ownership modes.
  6. Validate with python scripts/validate_harness_structure.py <root> and the repository's safe document checks, if known.
  7. Re-read generated conclusions against the ledger. Any unsupported statement becomes INFERRED or UNRESOLVED, not a polished fiction.

Allowed and forbidden changes

Allowed: new or explicitly approved Harness files and managed sections.

Forbidden: business code, dependency manifests, product-source rewrites, unapproved overwrites, feature reduction, invented stack or commands, and claims that an unrun check passed.

Failure handling

On missing sources, unreadable files, conflicts, or validation failure, preserve the ledger and report what is blocking. Roll back only untouched files created by this run. A second unchanged run must propose no churn; use harness-check to assess the resulting Harness.

Result format

Report source inventory, status counts, generated/updated/preserved files, conflicts, unresolved items, actual validation evidence, and manual acceptance steps. Recommend harness-prompt only when an actionable plan now exists.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.