Harness DevOps

harness-engineering-templates

A library of opinionated Claude Code harness templates for software engineering across four curated domains — web, data, DevOps, and mobile.

The premise (per Birgitta Böckeler, Anthropic engineering, METR, and the practitioner consensus of 2025–2026):

Agent = Model + Harness. Templates that lean on the model's good judgement fail their first incident. The harness — not the agent — is the contract.

The full reasoning lives in docs/explanation/why-harness.md.

---

Start here

Install from inside Claude Code — add the marketplace, then install the pack for your domain:

/plugin marketplace add Camilool8/harness-engineering-templates
/plugin install harness-web@harness-engineering      # or harness-data / harness-devops / harness-mobile
/harness-web:init                                    # pick a sub-domain

Installing any domain pack pulls in harness-base automatically (the four non-negotiable safety hooks). Full walk-through: docs/tutorials/getting-started.md. Plugin catalog and the HARNESS.toml opt-in flags: docs/reference/plugins.md.

Prefer committed .claude/ artifacts in your repo (audit-heavy or regulated teams)? The bash assembler is still supported as an eject path — see docs/reference/eject.md.

---

What this gives you

Five installable plugins — a shared harness-base plus four curated domain packs (harness-web, harness-data, harness-devops, harness-mobile):

• Four non-negotiable safety hooks — secret scanning, command guarding, append-only audit log, verification gate. Always on. See why.
• Opt-in discipline — TDD, eval-driven, two-key, and kill-switch hooks that arm only when you set a flag in .claude/HARNESS.toml; memory, progress-tracking, methodology, and orchestration guidance ship as auto-loading skills.
• Curated domain packs with sub-domains, addons, agent teams, and domain-specific gates (accessibility verify loop for web, unbounded-SQL block for data, plan-before-apply for devops, simulator-in-the-loop for mobile).
• Permissions stay yours — plugins ship the hooks (the enforcement contract) but never a permissions block; see docs/reference/recommended-permissions.md for an opt-in starting point.

Every default is the 2026 practitioner consensus.

---

Documentation

You want to…	Read
Install a pack for the first time	docs/tutorials/getting-started.md
Browse the plugin catalog + HARNESS.toml flags	docs/reference/plugins.md
Pick a domain pack and sub-domain	docs/how-to/pick-a-recipe.md
Use the bash assembler (eject path)	docs/reference/eject.md
Understand the design	docs/explanation/why-harness.md
Contribute a module, addon, or sub-domain	docs/how-to/ and CONTRIBUTING.md

The full documentation index, organised by Diátaxis quadrant: docs/README.md.

Three deep-reference essays (the why behind the design):

• docs/HARNESS_ENGINEERING.md — foundations, the Claude Code primitive reference, per-domain templates, cross-cutting concerns, universal anti-patterns.
• docs/METHODOLOGIES.md — SDD, TDD, BDD, ATDD, DDD, Agile, Waterfall, Lean, DevOps/SRE, Eval-Driven Development — each adapted to AI harnesses.
• docs/AGENT_ROLES.md — single-agent baseline, multi-agent topologies, the canonical role catalog, sub-agent design principles.

---

Repository layout

docs/                              user-facing documentation (start here)
  tutorials/                       learning-oriented walk-throughs
  how-to/                          problem-oriented recipes
  reference/                       schemas, CLI, catalogs, glossary
  explanation/                     understanding-oriented design notes
  HARNESS_ENGINEERING.md           deep reference
  METHODOLOGIES.md                 deep reference
  AGENT_ROLES.md                   deep reference