{
  "markdown": "# AI-Native Developer Experience Harness\n\n[![Listed on ClaudePluginHub](https://www.claudepluginhub.com/badge/jpantsjoha-join-the-team)](https://www.claudepluginhub.com/plugins/jpantsjoha-join-the-team?ref=badge)\n[![Agent Plugins 1.0.0](https://img.shields.io/badge/Agent%20Plugins-1.0.0-blue)](https://agent-plugins.org/specification)\n[![Agent Skills](https://img.shields.io/badge/Agent%20Skills-conformant-blue)](https://agentskills.io/specification)\n\n> **A team-project AI harness bootstrap that gives humans and agents a shared operating contract from day one, moving AI leverage from an individual “IC superhero” advantage to a repeatable team capability on an equal playing field.**\n\n> 📦 **Now listed on [Claude Plugin Hub](https://www.claudepluginhub.com/plugins/jpantsjoha-join-the-team)** —\n> install the harness as the `join-the-team` plugin for Claude Code, Codex, Kimi Code,\n> and Google Antigravity. [Jump to install](#install--two-ways-to-adopt).\n\nThis repository has gone through three real phases — and the progression tracks something\nhappening across the field.\n\nIt started as a **reference guide**: [`DEVELOPER_EXPERIENCE.md`](DEVELOPER_EXPERIENCE.md)\nand [`BOOTSTRAP.md`](BOOTSTRAP.md), something to read, walk through, and adapt. A concrete\nbaseline for teams getting serious about AI-native delivery — the operating model, the\nguardrails, spec-first discipline, the Adversarial Gate.\n\nThen it became **context for your agents**: `CLAUDE.md` and `GEMINI.md` here, plus the\nthin `AGENTS.md`/`CLAUDE.md`/`GEMINI.md` adapters the bootstrap generates — drop them\ninto a project and the agent picks up the operating contract at session start without a\nbriefing. The harness shifted from something humans read to something agents use.\n\nNow it ships as an **installable plugin** — `join-the-team` (current version in\n[CHANGELOG.md](CHANGELOG.md)). One install across\nClaude Code, Codex, Kimi, and Antigravity, and the full harness is live: 21 skills,\nsession-start orientation, slash commands, drift-checked in CI. The discipline travels\nwith the agent from day one, across every project, without anyone copying files or\nre-explaining the contract.\n\nThe point: this stopped being something you read and became something you install.\n\nAnd what you install is the **operating model** — that is the value proposition here.\nSkill libraries are everywhere now; what teams are missing is the contract: who holds\nauthority, how risk is tiered, what evidence binds a review, what \"done\" actually means\nwhen humans and agents ship together. The 21 skills are the capability layer that\nexecutes inside that contract — not the other way round.\n\nThe kernel stays model-, vendor-, and IDE-agnostic throughout. Thin platform adapters\nhandle discovery and invocation; the authority, risk, evidence, review, and completion\nsemantics are non-negotiable.\n\n> **Speed is easy. Safe speed is engineered.**\n\n> **This is deliberately opinionated.** Refine it for your team, technology, domain,\n> authority model, and definition of done. Keep one coherent shared contract while doing so.\n\n**New project?** [Bootstrap in fifteen minutes](BOOTSTRAP.md). **Existing agent setup?**\n[Two ways to adopt](#install--two-ways-to-adopt) below — by hand, or as a plugin.\n\n## Install — two ways to adopt\n\n**By hand** — clone the repo and share it with your agents. Copy `skills/` into\nyour project, keep the `CLAUDE.md`/`GEMINI.md`/`AGENTS.md` adapters at root, and your\nassistant picks up the contract at session start. [BOOTSTRAP.md](BOOTSTRAP.md) is the\nfifteen-minute walkthrough.\n\n```bash\ngit clone https://github.com/jpantsjoha/ai-native-developer-experience\n```\n\n**As a plugin** — the harness ships as `join-the-team`: one canonical skill\nset, thin per-harness adapters, drift-checked in CI. It is listed on\n[Claude Plugin Hub](https://www.claudepluginhub.com/plugins/jpantsjoha-join-the-team);\ninstall straight from this repository into whichever coding assistant you run:\n\n**Claude Code**\n\n```text\n/plugin marketplace add jpantsjoha/ai-native-developer-experience\n/plugin install join-the-team@join-the-team-marketplace\n```\n\n**Kimi Code**\n\n```text\n/plugins install https://github.com/jpantsjoha/ai-native-developer-experience\n```\n\n**Antigravity (Gemini)**\n\n```bash\nagy plugin install https://github.com/jpantsjoha/ai-native-developer-experience\n```\n\n**Codex** — no install command needed: Codex reads `AGENTS.md` as its always-on adapter and\ndiscovers skills from `.agents/skills/`. The by-hand clone above is the install; if you are\nvendoring rather than cloning, put the directory at `.agents/skills/` (or symlink it there)\nso Codex finds it.\n\nFull per-harness detail (session-start hooks, verification, update path):\n[Claude Code](docs/install/claude.md) · [Kimi Code](docs/install/kimi.md) ·\n[Codex](docs/install/codex.md) · [Antigravity](docs/install/antigravity.md)\n\nAntigravity is a first-class surface: this harness was built and battle-tested on\nGoogle Cloud's agent stack, and ships cloud-expert guardrails (`gcp-expert`,\n`aws-expert`, `azure-expert`, `alibaba-expert`) plus `adk-expert` and\n`mcp-server-scaffold` alongside the vendor-neutral contract.\n\nThe plugin composes with — never duplicates — companion skill plugins;\nsee [INTEGRATIONS.md](INTEGRATIONS.md).\n\n### Verify the install\n\nAfter installing, confirm the harness is actually live rather than merely present:\n\n```text\nwhich skills are available?\n```\n\nYou should see `using-the-harness` plus the cloud and gate skills (`gcp-expert`,\n`adversarial-gate`, `pr-reviewer`). Then run the built-in validator against a\nbootstrapped repository:\n\n```text\n/join-the-team:validate\n```\n\nExpect `PASS operating-model validation`. A `seed` profile may pass with warnings — that\nis expected before the profile is promoted to `active`.\n\n## Usage\n\nThe harness is not a set of commands you memorise. Skills trigger on intent; three slash\ncommands cover the lifecycle.\n\n**Three commands, in order:**\n\n```text\n/join-the-team:bootstrap    # seed the operating model (dry-run first, never overwrites)\n/join-the-team:init         # record the humans: roles, authority, escalation channels\n/join-the-team:validate     # check adapter drift, adoption state, evidence binding\n```\n\n**Typical sessions** — say the intent, and the router picks the skill:\n\n| You say | What fires | What you get |\n| --- | --- | --- |\n| \"Design the ingestion service\" | `spec-first-delivery` → `the-architect` | Spec and ADR *before* code |\n| \"Is this safe to ship?\" | `release-readiness` | Go/no-go against failure modes and rollback |\n| \"Review this PR\" | `pr-reviewer` | Structured verdict with findings |\n| \"How would this break?\" | `adversarial-gate` | Red-team pass argued against your own approach |\n| \"We're on GCP, multi-tenant\" | `gcp-expert` + `governance-guardrail` | IAM, residency and policy guardrails |\n| \"Where do things stand?\" | `sitrep` | Status, blockers, next actions |\n\n**Worked example — a risk-touching change:**\n\n```text\nYou:   We need to let tenants export their own audit logs.\nAgent: [spec-first-delivery] Spec before code. Who is the actor, what is out of scope?\n       [governance-guardrail] Multi-tenant export is R2 — data-boundary check required.\n       [the-architect] ADR-00N drafted: signed URLs vs streamed export, trade-offs recorded.\n       [adversarial-gate] How would I break this? Tenant A requesting tenant B's logs.\n       → Named human approves the R2 classification before implementation starts.\n```\n\nThe point is the routing, not the vocabulary: you describe intent, the contract decides\nwhich gate applies.\n\n## Configuration\n\nThe harness works unconfigured. Three seams are worth knowing:\n\n| Seam | File | What it controls |\n| --- | --- | --- |\n| **Operating contract** | `CLAUDE.md` / `GEMINI.md` / `AGENTS.md` | Always-on rules the agent loads each session |\n| **Project profile** | `PROJECT-OPERATING-PROFILE.md` (generated by `bootstrap`) | Risk tiers, named authority, escalation, definition of done |\n| **Data seams** | `.agents/mcp_config.json` | Governed MCP access — a **template**, not live config |\n\nAdjust the risk tiers and authority model to your team; keep one coherent contract while\nyou do. The MCP file ships as an example: copy an entry, rename it, point it at a source\nyou control. See\n[the governed-data-seam pattern](DEVELOPER_EXPERIENCE.md#the-governed-data-seam-pattern).\n\n## Troubleshooting\n\n| Symptom | Cause | Fix |\n| --- | --- | --- |\n| Skills discovered but no orientation at session start | The session-start hook needs `bash` on `PATH` | Install `bash`; without it the plugin degrades to discovery without injection |\n| Antigravity reports `hooks: skipped (not found)` | The root `hooks.json` is missing from the install | Reinstall; `agy plugin install` must report `hooks: 1 processed` |\n| Codex-installed copy has no `.agents/skills/` | Codex's install cache flattens symlinks | Expected — all 21 skills are at `skills/`; use `.agents/skills/` only in a repo checkout |\n| `/join-the-team:validate` warns about unresolved fields | Profile is still a `seed` | Fine for R0/R1 work; resolve placeholders and promote to `active` before R2/R3 |\n| Slash commands missing after install | Client not restarted | Restart the client; Claude Code applies plugin updates on restart |\n| Agent ignores the contract mid-session | Context drift after a long session | `/clear`, then let the session-start hook re-inject the orientation skill |\n\nStill stuck? Open an [issue](https://github.com/jpantsjoha/ai-native-developer-experience/issues)\nwith the client, version (`plugin.json`), and the output of `make spec-conformance`.\n\n---\n\n## Is this the same as BMAD?\n\nNo — and the difference is the point. [BMAD](https://docs.bmad-method.org/) is a\nmethodology engine: it drives an idea through phased workflows — analysis, planning,\nsolutioning, implementation — with AI personas playing Analyst, PM, Architect, Developer.\nIt answers *what to do next*. `join-the-team` answers *who is accountable, and what is\nallowed* — the contract the work runs **inside**: risk tiers, named human authority,\nevidence bound to an exact commit, a definition of done that means the checks pass. They\nscale on different axes — BMAD scales *ceremony by size*, this harness scales *gates by\nrisk*. So they compose rather than compete: run BMAD as the workflow engine, and let\n`join-the-team` keep a mixed human-and-agent team coherent while it does. Reach for it the\nmoment *\"who decided this, and on what evidence?\"* stops being obvious. The workflow you\ncan borrow anywhere — the contract is the part teams keep missing.\n\n---\n\n## What it looks like\n\nInstall the plugin, and at session start the `using-the-harness` skill grounds the agent\nin the contract before any work. Ask it what the harness changes, and it answers in the\nterms that matter — where the system accelerates, and where it stops:\n\n![join-the-team in a Claude Code session: the using-the-harness skill explaining that the plugin delivers coherence and auditability rather than speed, contrasting work with and without the harness, and showing how a decision routes through the adversarial gate, domain-validator, and pr-reviewer.](docs/images/harness-in-action.png)\n\n> **Coherence and auditability — not speed.** The value is visible when the system knows\n> where to accelerate *and* where to stop: a risk-touching decision triggers the\n> adversarial gate, evidence is re-verified before it becomes actionable, and the\n> decision is captured as a receipt. Same contract for the humans and the agents.\n\n---\n\n## The Hybrid Human-AI Squad Model & Workflow\n\nInstalling the harness gets you the skills. What it doesn't give you is the operating\nmodel — who decides, who executes, what counts as done, where the escalation circuit\nbreakers sit. This is that model.\n\nMoving from an individual \"copilot user\" to a team delivering value requires shifting from ad-hoc prompting to a **governed value stream**. In this model, **AI agents and harness skills scale execution velocity**, while **named human Subject Matter Experts (SMEs) retain non-delegable accountability** for decisions, governance, and production state.\n\n```mermaid\n%%{init: {'sequence': {'boxMargin': 15, 'messageMargin': 30}, 'themeVariables': {'actorLineColor': '#000000'}}}%%\nsequenceDiagram\n    actor PO as Product Owner (Human SME)\n    participant Orch as delivery-orchestrator (AI Workflow)\n    participant SpecArch as spec-first & the-architect (AI Skills)\n    actor SME as Human SME Roster (Architect / Data / Delivery Lead)\n    participant Lanes as Agent Execution Lanes (AI Agents & Hooks)\n    participant Gates as domain-validator & adversarial-gate (AI Gates)\n    actor Ops as Operations / SRE (Human SME)\n\n    rect rgb(195, 210, 255)\n        note over PO,SpecArch: Phase 1: Intake & Acceptance Contracts (Human Intent & AI Spec)\n        PO->>Orch: 1. Submit Epic / Feature Intent\n        Orch->>SpecArch: 2. Route Work & Draft Contracts\n        SpecArch-->>Orch: 3. Return Acceptance Contract & ADRs\n    end\n\n    rect rgb(255, 185, 205)\n        note over Orch,SME: Phase 2: Risk Governance & Escalation Circuit Breaker (Human SME Authority)\n        Orch->>Orch: Classify Risk Tier (R0-R3) & Check Evidence\n        alt Missing Evidence or High-Risk (R2/R3)\n            Orch-->>SME: LANE HALT: Tag Accountable Human SME\n            SME->>Orch: Approve Authority Grant & Sign Decision Record (ADR)\n        end\n    end\n\n    rect rgb(160, 230, 175)\n        note over Orch,Gates: Phase 3: Agent Task Execution & Red-Team Pass (AI Execution Lanes)\n        Orch->>Lanes: 4. Dispatch Discrete Tasks to Mutating Lanes (R0/R1 Scoped)\n        Lanes->>Gates: 5. Execute Code & Run Red-Team Pass (\"How would I break this?\")\n        Gates-->>Lanes: 6. Pass Verification & Bind Exact-Candidate SHA\n    end\n\n    rect rgb(160, 215, 245)\n        note over Lanes,Ops: Phase 4: Release Readiness & Derived Receipts (Human Sign-off & Audit)\n        Lanes->>Ops: 7. Submit Candidate Release Check (pr-reviewer & release-readiness)\n        Ops-->>Lanes: 8. Authorise Production Deployment\n        Lanes-->>PO: 9. Deploy & Emit Derived Status (sitrep)\n    end\n```\n\n### Accountabilities: Human Squad vs. AI Harness\n\nSkills supply capability; **named humans supply authority**.\n\n| Delivery Stage | Primary AI Skill / Harness Primitive | AI Agent Capability | Accountable Human SME |\n| :--- | :--- | :--- | :--- |\n| **Requirements & Scope** | `spec-first-delivery` | Drafts acceptance contract & spec | **Product Owner** |\n| **Architecture & ADRs** | `the-architect`, `gcp-expert` / `aws-expert` / `azure-expert` | Drafts ADRs & validates vendor constraints | **Lead Architect / Head of Eng** |\n| **Data & Access Seams** | `mcp-server-scaffold` | Queries governed MCP data seams | **Head of Data / Security** |\n| **Risk & Authority** | `using-the-harness` | Classifies risk tier (R0–R3) | **Delivery Manager / Lead** |\n| **Verification & Red-Teaming** | `adversarial-gate`, `make check` | Runs red-team checks & test suite | **Lead Developer** |\n| **Release & Rollback** | `release-readiness` | Validates deployment readiness | **Operations / SRE** |\n| **Status & Receipts** | `sitrep` | Synthesises status from git artifacts | **Delivery Manager** |\n\n### Core Operating Invariants\n\n- **Capabilities scale velocity; Authority remains human**: AI agents draft code, specs, and execution plans, but named humans approve decisions at declared risk tiers.\n- **The Escalation Circuit Breaker (\"Silence never converts to permission\")**: When evidence is missing or work touches R2/R3 risk, the execution lane halts and tags the human owner.\n- **Receipts over polish**: Done means checks pass (`make check`), evidence manifests bind to exact commit SHAs, and status is derived from artifacts, not prose.\n\n---\n\n## Licence\n\nThis repository is open source under the [Apache License, Version 2.0](LICENSE) — free for personal and commercial use, modification, and redistribution.\n\nThe plugin's data-handling statement is available in the [Privacy Policy](docs/PRIVACY.md).\n\nAttribution is part of the deal: derivative works must carry the [NOTICE](NOTICE)\nfile (Apache-2.0 §4(d)), which credits the author and this project's origin. The repository contains no usage telemetry; the licence cannot identify silent use. Stars, feedback, and voluntary adoption notes are welcome evidence that the harness is useful, but they are not licence conditions.\n\n## This repo is about\n\n- A **living DevEx harness** for AI-augmented development.\n- A record of **what worked** in firsthand delivery experience.\n- A set of **minimum viable guardrails** for agent-driven workflows.\n- An opinionated baseline designed for team refinement.\n\nHyper-personalised workflows are inevitable; this is one of many. Shared contracts,\nvalidation, and discipline remain essential.\n\n## Where the harness came from\n\nI’ve been building and writing under the banner of **#HarnessEngineering** for a while now — the idea that the model is the easy, commoditised part - relatively speaking, and the durable engineering lives in the scaffolding you wrap around it. So the rule files, the tools and MCP servers, the sandboxes, the orchestration, the hooks, the evals. This repo is the firsthand version of that argument — a reflection of over a year working hands-on across a variety of agent systems, coding copilots, and orchestrated multi-agent delivery.\n\nIf you want the narrative rather than the code, the write-ups that unpack this harness live here:\n\n- **Part 2 — [Harness Engineering with Google Antigravity](https://medium.com/google-cloud/the-ai-native-developer-experience-part-2-harness-engineering-with-google-antigravity-7fb72dab243f)** (Medium, Google Cloud Community) — one harness across three surfaces: the IDE, the Agent Manager, the CLI.\n- **Companion — [Supercharging Your Harness: Skills, Rules and MCP with Google Antigravity](https://itnext.io/supercharging-your-harness-skills-rules-and-mcp-with-google-antigravity-d2142e61c4fd)** (ITNEXT) — how the Skills, Rules and MCP primitives in this repo actually come together to make the harness invocable.\n\nThe broader industry is converging on similar language. One useful marker is Google’s\n2026 paper [“The New SDLC With Vibe Coding: From ad-hoc prompting to Agentic\nEngineering”](https://www.kaggle.com/whitepaper-the-new-SDLC-with-vibe-coding) by Addy Osmani, Shubham Saboo, and Sokratis Kartakis.\n\nA few related themes and external signals line up with what this repository has been\nsaying from the field:\n\n- **The harness can dominate the experience.** “10% model / 90% harness” is a useful\n  engineering heuristic, not a universal measured ratio. The practical point is to debug\n  tools, context, rules, permissions, and feedback loops as first-class system components.\n- **Many apparent model failures are harness failures.** Missing tools, vague rules,\n  absent guardrails, poor context, and weak validation are common, actionable causes. This\n  is a field observation, not a claim that every failure has the same root cause.\n- **The harness effect can be measured.** LangChain reported improving Deep Agents from\n  52.8 to 66.5 on Terminal Bench 2.0 through harness changes, moving from outside the Top\n  30 to the Top 5 at that time. Treat the result and rank as a historical experiment, not\n  a permanent benchmark fact. See [LangChain’s experiment](https://www.langchain.com/blog/improving-deep-agents-with-harness-engineering)\n  and the [current Terminal Bench leaderboard](https://www.tbench.ai/leaderboard/terminal-bench/2.0).\n- **Adoption is widespread, but measurements differ.** Surveys often mix AI tools,\n  coding assistants, and agents, so this repo does not turn tool-use percentages into an\n  “agent adoption” or “AI-generated code” claim. See the [JetBrains 2025 ecosystem\n  report](https://blog.jetbrains.com/research/2025/10/state-of-developer-ecosystem-2025/) and [Stack Overflow 2025 survey](https://survey.stackoverflow.co/2025/ai).\n- **The role is shifting from syntax to intent** — from writing code to specifying, verifying, and directing — with “intent as the new interface” as the destination.\n\nBasically, **read that report.**\n\n## Patterns worth borrowing: agent-skills convergence\n\nAddy Osmani followed the paper with a practical artifact — **[agent-skills](https://github.com/addyosmani/agent-skills)** (MIT), 24 SKILL.md workflows encoding SDLC discipline for coding agents. I reviewed the lot against this harness. Most of it my setup (or your agent CLI of choice) already does natively. Four patterns are genuinely worth lifting, and they slot straight into the harness thinking above:\n\n1. **Anti-rationalization tables.** Every skill ships a table of the excuses an agent makes to skip a step — paired with the rebuttal. This is a harness primitive I had not formalised: my gates ban bad *output*; this pattern pre-empts bad *reasoning* before the output exists. If you maintain your own skills, add one of these tables to each. Cheap to write, compounds fast.\n2. **Doubt-driven development.** Adversarial in-flight review of high-stakes decisions — the agent must argue against its own approach before proceeding. I've been running this as the **Adversarial Gate** (\"how would i break this?\") since the start of this harness. Good to see the industry converge on the same move. If you only borrow one behavioural pattern, borrow this one.\n3. **A meta-router skill.** As a skill library grows, the agent needs explicit routing.\n   This harness implements that capability in `delivery-orchestrator`; the operating\n   profile then maps capability names to whichever invocation syntax the team uses.\n4. **Exit criteria over aspirational guidance.** The repo's quiet philosophy: process over prose. A skill that says \"ensure quality\" is decoration; a skill that says \"done means these three checks pass\" is a harness. Same discriminator i keep landing on everywhere: receipts, not polish.\n\nBorrow the patterns. As ever — your mileage may vary.\n\n## Standards conformance\n\n**`join-the-team` is compliant with [Agent Plugins 1.0.0](https://agent-plugins.org/specification)**\n— the open, vendor-neutral packaging standard\n[announced by Google](https://developers.googleblog.com/agent-plugins-package-your-skills-tools-and-more/)\nand stewarded by a Technical Steering Committee spanning Amazon, Cursor, Google, Microsoft,\nOpenAI and Vercel. Its skills comply with\n[Agent Skills](https://agentskills.io/specification).\n\nThat is a gate, not a badge — run it yourself:\n\n```bash\nmake spec-conformance\n```\n\nThe check is standard-library Python, runs offline, and reports as its own CI job. It is\nbacked by 51 negative fixtures in `tests/test_plugin_packaging.py`, because a validator that\nonly ever passes is decoration.\n\n**Scope of compliance, stated precisely.** The standard defines exactly **two** component\ntypes — skills (`skills/`) and MCP servers (`mcp.json`). Commands, hooks, agents, rules and\nLSP servers are explicitly **outside v1** (§7), so this plugin's slash commands and\nsession-start hooks are client-specific concerns, not conformance surface. They are declared\nas **manifest extensions** (§8.1), whose contents the standard assigns no meaning to and\nwhich conformant clients must ignore for namespaces they do not implement. This package does\n**not** claim `.claude-plugin/` or `.kimi-plugin/` as §8.2 *directory* extensions — those\nmust be named after the namespace itself; ours are ordinary top-level directories, which the\nstandard treats as non-errors.\n\n**What conformance buys you.** The root [`plugin.json`](plugin.json) is one portable entry\npoint that any conformant client can load. The four vendor manifests —\n`.claude-plugin/`, `.kimi-plugin/`, `gemini-extension.json`, and the marketplace entry —\nremain as client-specific projections, declared through the standard's `extensions` block\nrather than left for a client to guess at. A harness that sells one shared contract should\nnot ship as a vendor fork.\n\n**What the gate actually checks** — spec-required rules first, local hygiene marked as such:\n\n| Surface | Rule | Source |\n| --- | --- | --- |\n| Root manifest | `$schema` const, name pattern, no key outside the ten the schema permits, `author` shape | Agent Plugins 1.0.0 |\n| Skills | Name pattern and 64-char cap, frontmatter name matches directory, non-empty description within 1024 chars | Agent Skills |\n| `skills/` fixed location | Real directory, not a symlink; `.agents/skills` alias stays relative and in-root | Spec (path safety) + local |\n| `extensions` | Reverse-domain namespace keys; every declared plugin-relative path exists on disk | §8.1 + local |\n| `mcp.json` | Closed transport union, reserved env vars, `cwd` rooting — enforced if the file is ever added | Agent Plugins 1.0.0 |\n| Hook manifests | `hooks/hooks.json` (Claude Code) and root `hooks.json` (Antigravity) exist and are identical | **Local only** — hooks are outside the spec |\n| Six manifests | Name and version agree across root, vendor and marketplace manifests | **Local only** |\n\n**Two constraints worth knowing.** Canonical skills live in `skills/` — a real directory at\nthe standard's fixed discovery location — with `.agents/skills/` as a relative symlink for\nrunners that discover there natively (Codex, Kimi). That layout is the result of live install\ntesting, not theory: an earlier build had it the other way round, and **Codex's installer\ndropped the symlink**, leaving a conformant client with zero skills at the fixed location.\nThe real directory now sits where the standard looks. And this package ships **no root\n`mcp.json`**: it is optional in the standard, and\n`.agents/mcp_config.json` is a teaching template naming example servers no client should ever\nspawn. [ADR-002](architecture/decisions/ADR-002-agent-plugins-spec-conformance.md) records\nboth decisions and the trade-offs behind them.\n\n## What’s inside this repo\n\n- **[BOOTSTRAP.md](BOOTSTRAP.md)**\n  The drop-in, fifteen-minute path for a new team project.\n- **[DEVELOPER_EXPERIENCE.md](DEVELOPER_EXPERIENCE.md)** (DX-001)\n  The main guide covering guardrails, workflows, validation, spec-driven delivery, and AI-agent integration — the mechanics.\n- **[Agent Skills Library](skills/README.md)**\n  Twenty-one tracked capabilities for orchestration, architecture, specification, validation,\n  review, release readiness, governance, plugin submission, status, cost, and selected\n  platform work.\n- **[Plugin Submission](skills/plugin-submission/SKILL.md)**\n  The policy-backed directory and marketplace listing gate; its\n  [capability specification](docs/PLUGIN-SUBMISSION.md) defines the external-send\n  confirmation and receipt contract.\n- **[Operating Model Bootstrap](skills/operating-model-bootstrap/SKILL.md)**\n  A reusable released manual, project profile, checkpoint, evidence manifest, and thin\n  agent-surface adapters for assigning authority, isolating parallel human/agent lanes,\n  binding review to an exact candidate, and carrying work through delivery, observation,\n  and honest completion.\n- **[Companion Plugins](INTEGRATIONS.md)**\n  The evaluated companion-plugin map: agent-craft lanes (TDD methodology, simplicity\n  discipline, frontend design) route to installed sister plugins; the team contract\n  stays canonical here. Reference, never vendor.\n- **[Team Workflow](docs/WORKFLOW.md)**\n  The skills dependency diagram, the requirement → ADR → ticket → evidence → status\n  traceability chain, and the accountability model: how skills route work while named\n  humans hold authority.\n- **[CHANGELOG.md](CHANGELOG.md)**\n  Material changes, narrow public-source attribution, and known usage limitations.\n\nThis repository is expected to **evolve** as tools, models, and workflows change. Dated\nfacts and prices are snapshots; verify them before making a current decision.\n\n## Projects that helped shape this harness\n\nThese ideas were not written in isolation — they were forged while building, shipping, breaking, and iterating on real systems using AI-assisted and agent-driven workflows. (Besides my own delivery experience in the field)\n\n### My Hackathons and Builds\n\n- **[Devpost — project & hackathon portfolio](https://devpost.com/jpantsjoha/achievements)**\n  My running track record of things I’ve designed, built, and shipped — hackathon entries, prototypes, and production tools. The proof-of-work behind the opinions in this harness.\n\n### Chrome Apps\n\n- **[Simple Focus Mode – Chrome Extension](https://github.com/jpantsjoha/simple-focus-chromeExt)**\n  A minimalist Pomodoro-style productivity extension focused on intentional work and reduced distraction.\n  Built as a fast-feedback experiment in shipping, UX constraints, and iterative delivery.\n  Related write-up:\n\n  - https://medium.com/devops-dudes/simple-focus-mode-boost-your-productivity-with-my-chrome-extension-3f5cf0f7b843\n- **[C4X – C4 Model Diagrams for VS Code](https://marketplace.visualstudio.com/items?itemName=jpantsjoha.c4x)**\n  A VS Code extension for authoring and visualising C4 architecture diagrams with AI-assisted generation and live preview.\n  Used to explore developer experience, documentation-as-code, and AI-assisted design workflows.\n  Additional links:\n\n  - Open VSX: https://open-vsx.org/extension/jpantsjoha/c4x\n  - Source: https://github.com/jpantsjoha/c4x-vscode-extension\n  - Build story: https://medium.com/google-cloud/how-i-built-the-c4x-antigravity-ide-extension-with-googles-gemini-3-6feb74f8a4b2\n\n### 🧠 Slack + Cloud AI\n\n- **[BriefOps – Slack + Google Cloud AI Summarisation](https://github.com/jpantsjoha/briefops-public)**\n  A Slack application leveraging Google Cloud and Vertex AI to summarise conversations, documents, and shared context.\n  Used to explore agent-assisted knowledge extraction, permissions, governance, and delivery guardrails in a real SaaS workflow.\n  Related article:\n  - https://medium.com/google-cloud/slack-googlecloud-briefops-streamlining-slack-comms-with-gcp-ai-powered-summarisation-ec2151672731\n\n## My Thoughts on this Agentic Ways of Working\n\n- [The New Ways of Working: Leading with Agent-Powered Hybrid Teams](https://www.cognizant.com/uk/en/insights/blog/articles/the-new-ways-of-working-leading-with-agent-powered-hybrid-teams)\n  The published thesis behind this harness, with reflection on hybrid human-and-agent teams, the harness over the model, machine-readable contracts, and orchestration as the new core skill, and more\n- **Medium**: https://jaroslav-pantsjoha.medium.com  My GDE and Technical write-ups\n  Typical subjects i cover:\n\n  - Developer experience\n  - Platform engineering\n  - Cloud & AI delivery\n  - Agentic workflows\n  - Applied AI\n\nRelevant posts will be cross-linked here as this harness evolves.\n\n- [Supercharging Your Harness: Skills, Rules and MCP with Google Antigravity](https://itnext.io/supercharging-your-harness-skills-rules-and-mcp-with-google-antigravity-d2142e61c4fd) (ITNEXT) — how the Skills, Rules and MCP primitives come together to make the harness ship safe, consistent output.\n\n## About the author\n\nCreated and maintained by **Jaroslav Pantsjoha** — Technical Director and Enterprise Agent Solution Architect, Google Developer Expert (Google Cloud), speaker, and consulting thought leader on AI systems and adoption. Prolific builder, learner, and the author of the **\"Ways of Working and AI Adoption at the Enterprise\"** series and two books:\n\n- *Building the Agentic Enterprise on Google Cloud* (Packt) — a practical field guide to designing, deploying, and operating agentic AI systems.\n- *Mastering Multi-Agent Systems on Google Cloud* (AVA Publishing, Co-Author with Anupam Phoghat) — build, deploy, and operate production agentic AI with ADK, Vertex AI, and the complete GCP stack.\n\n- LinkedIn: [uk.linkedin.com/in/johas](https://uk.linkedin.com/in/johas)\n- Google Developer Expert: [me.developers.google.com/u/jpantsjoha](https://me.developers.google.com/u/jpantsjoha)\n\nAttribution requirements for copies and adaptations live in [NOTICE](NOTICE). The Adversarial Gate name and \"how would I break this?\" framing are his contribution to the #HarnessEngineering body of work.\n\n## Status\n\nThis is an active, evolving project.\nExpect revisions, additions, and corrections as tools and practices mature.\n\nFeedback, discussion, and constructive disagreement are welcome.\n",
  "bytes": 32278,
  "sha": "1010b46da7988a9e327f7955629a760474fe125a2302e94842376f5593ba1cf5",
  "repo_slug": "jpantsjoha/ai-native-developer-experience",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_jpantsjoha_ai_native_developer_experienc_e360918d/readme"
}