By rodhappi
The minimum set of primitives any AI system needs to operate with integrity. Provides 6 structural primitives (Identity, Context, Honesty, Memory, Learning, Observability) enforced through Claude Code's native mechanisms.
4-phase learning synthesis between sessions. Processes staged signals, checks for drift, and updates persistent memory. Use between sessions to process learnings.
Show currently loaded context, declared absences, and context budget utilization. Use to check what context is loaded and what is missing.
Update an existing domain harness while enforcing layered mutability. L1 elements are immutable, L2 requires confirmation, L3 is freely modifiable. Use when modifying harness configuration.
Verify all 6 primitives are intact, coherent, and self-contained. Runs 36-point validation (primitives + self-containment). Use after creating or modifying a harness.
Create a new domain-specific harness. Guided discovery (13 questions) produces a complete Claude Code project with all 6 structural primitives inherited from the meta-harness. Use when starting a new AI system project.
Matches all tools
Hooks run on every tool call, not just specific ones
Executes bash commands
Hook triggers when Bash tool is used
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Modifies files
Hook triggers on file write and edit operations
Modifies files
Hook triggers on file write and edit operations
A Claude Code plugin that provides the structural foundation any AI system needs to operate with integrity. Install it, run /meta-harness:new-harness, answer 13 questions about your domain, and get a fully configured system with hallucination prevention, memory management, learning governance, and observability — all inherited, all enforced, all structural.
Most AI systems fail in predictable ways: they hallucinate when they lack knowledge, they drift from their purpose over long sessions, they learn the wrong lessons from user feedback, and their guardrails can be prompt-injected away.
These failures share a root cause: governance is instructional, not structural. Telling a model "be honest" is an instruction. Gating every tool call through an external confidence check is a structure. Instructions can be forgotten, reasoned around, or compressed away. Structures cannot.
The meta-harness solves this by providing 6 structural primitives — enforced through Claude Code's native mechanisms (hooks, rules, skills, subagents) — that every domain system inherits without exception. You build your sales agent, research assistant, or code reviewer on top of these guarantees. You don't have to think about hallucination prevention or identity drift. It's already done.
Every system instantiated from the meta-harness inherits all six. Extension is allowed; removal is forbidden.
| Primitive | Question It Answers | How It's Enforced |
|---|---|---|
| Identity | Who am I? | CLAUDE.md + rules/identity.md — character-driven (virtue ethics), not rule-driven |
| Context Contract | What do I know before acting? | SessionStart hook — loads context, verifies completeness, declares absences |
| Honesty Protocol | What do I declare about my limits? | PreToolUse hook — confidence gate before every tool call, 5 types of not-knowing |
| Memory Protocol | What do I remember between sessions? | PreCompact hook + file-based hot/warm/cold tiers + WIP ledger — assumes interruption by default, survives both compaction and API failures |
| Learning Protocol | What do I extract from what I did? | Stop hook + /consolidate — stages signals during sessions, consolidates between them |
| Observability | What is happening in the system? | PostToolUse + SessionEnd hooks — loop detection, metrics, external governance |
| Concern | Structural Mechanism |
|---|---|
| Hallucination | PreToolUse hook gates every tool call on confidence. Absence declarations prevent speculation on missing info. Five distinct types of not-knowing with calibrated language. |
| Laziness | Character-driven identity (intellectual rigor as a functional virtue). Anti-epistemic cowardice rule. Conflict resolution: honest > helpful. Loop detection. |
| Memory loss | Hot/warm/cold tiers. PreCompact safety net persists state before compression. Assume-interruption principle. |
| Identity drift | CLAUDE.md survives compaction (re-read from disk). Drift-checker subagent validates learnings against identity. |
| Governance bypass | Hooks are shell scripts outside the reasoning loop. They fire whether the model wants them to or not. |
| Guarantee removal | /evolve-harness enforces layered mutability — L1 elements are immutable, L2 requires confirmation, L3 is free. |
| Interruption loss | WIP ledger doctrine: multi-step tasks persist progress to .claude/memory/wip.json after each step. API failures and crashes do not lose in-flight state. |
The meta-harness is a scaffolding generator, not a runtime framework. A generated domain harness is a self-contained Claude Code project that runs without the plugin being installed. Understanding this distinction is important:
| Meta-Harness Plugin (this repo) | Domain Harness (output of /meta-harness:new-harness) | |
|---|---|---|
| What it is | The scaffold. Source-of-truth templates, rules, hooks, schemas, operational skills. | The product. A self-contained Claude Code project with its own copies of everything it needs. |
| When it matters | Needed to create a harness. | Needed to run a harness. |
| Runtime dependency | None — the harness has no reference to ${CLAUDE_PLUGIN_ROOT}. | N/A — the harness IS the runtime. |
| How it changes | Manually: edit files, bump version, git commit. | Via /evolve-harness (copied into the harness) with L1/L2/L3 enforcement. |
| How updates propagate | Via dated patches in primitives/patches/ that operators opt into. | Via /evolve-harness for config changes, or by applying patches for structural upgrades. |
| Who changes it | The meta-harness maintainer. | The domain harness operator. |
npx claudepluginhub rodhappi/meta-harness-plugin --plugin meta-harnessUltra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Multi-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Curate auto-memory, promote learnings to CLAUDE.md and rules, extract proven patterns into reusable skills.
Memory compression system for Claude Code - persist context across sessions
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns