metaharness 0.1.0

package/templates/vertical_exotic/.claude/settings.json.tmpl

@@ -0,0 +1,44 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(npx {{name}}*)",
+      "mcp__{{name}}__*",
+      "mcp__evolution_log__*",
+      "mcp__federation__*"
+    ],
+    "deny": [
+      "Read(./.env)",
+      "Read(./.env.*)",
+      "Bash(rm -rf*)"
+    ]
+  },
+  "mcpServers": {
+    "{{name}}": {
+      "command": "npx",
+      "args": [
+        "-y",
+        "{{name}}@latest",
+        "mcp",
+        "start"
+      ]
+    },
+    "evolution_log": {
+      "command": "npx",
+      "args": [
+        "-y",
+        "{{name}}@latest",
+        "mcp",
+        "evolution"
+      ]
+    },
+    "federation": {
+      "command": "npx",
+      "args": [
+        "-y",
+        "{{name}}@latest",
+        "mcp",
+        "federate"
+      ]
+    }
+  }
+}

package/templates/vertical_exotic/.claude/skills/evolve/SKILL.md.tmpl

@@ -0,0 +1,14 @@
+---
+name: evolve
+description: "Run one safe self-improvement cycle: hypothesize → experiment → record → (maybe) federate."
+---
+
+# evolve
+
+Run one evolution cycle.
+
+1. Hypothesizer reads the evolution log and proposes a falsifiable change with a metric.
+2. Experimenter tests it in a sandbox and records a signed kept/killed result.
+3. Federator shares it to peers only if witness-signed and reproduced.
+
+Guard against Goodharting the metric. See ADR-014 (self-evolution + federation).

package/templates/vertical_exotic/.claude/skills/memory-inspect/SKILL.md.tmpl

@@ -0,0 +1,14 @@
+---
+name: memory-inspect
+description: Search and inspect the harness memory namespace (HNSW + emergent-time decay).
+---
+
+# memory-inspect
+
+Inspect what the harness has learned.
+
+- `search <query>` — semantic nearest-neighbour over the namespace
+- `list` — recent patterns with decay weight
+- `forget <id>` — evict a pattern
+
+Use this before planning so the harness reuses prior trajectories instead of starting cold.

package/templates/vertical_exotic/CLAUDE.md.tmpl

@@ -0,0 +1,31 @@
+# {{name}}
+
+{{description}}
+
+> Exotic / Self-Evolving harness · domain: `exotic/self-evolution`. Generated with [create-agent-harness](https://github.com/ruvnet/agent-harness-generator).
+
+## Behavioral rules
+
+- Use the harness's MCP tools (`mcp__{{name}}__*`) for orchestration
+- Memory and routing are handled by the kernel — you don't need to learn them
+- Defer destructive operations to the user
+
+## Agents
+
+| Agent | Tier | Role |
+|---|---|---|
+| `hypothesizer` | opus | Proposes a falsifiable self-improvement. |
+| `experimenter` | opus | Tests the hypothesis safely and records it. |
+| `federator` | sonnet | Shares vetted improvements across instances. |
+## Skills
+
+- `/memory-inspect` — Search and inspect the harness memory namespace (HNSW + emergent-time decay).
+- `/evolve` — Run one safe self-improvement cycle: hypothesize → experiment → record → (maybe) federate.
+
+## Commands
+
+- `doctor` — Health-check the harness: kernel load, MCP wiring, memory backend, host adapter.
+
+## Architecture
+
+This harness uses [@ruflo/kernel](https://www.npmjs.com/package/@ruflo/kernel) — a Rust-compiled WASM module with a NAPI-RS native fallback — so the same code runs identically on every platform.

package/templates/vertical_exotic/README.md.tmpl

@@ -0,0 +1,29 @@
+# {{name}}
+
+{{description}}
+
+> **Exotic / Self-Evolving** — Hypothesizer → experimenter → federator over a witness-signed evolution log (ADR-014).
+>
+> Generated with [`create-agent-harness`](https://github.com/ruvnet/agent-harness-generator). WASM kernel, multi-host support, witness-signed releases.
+
+## Install
+
+```bash
+npm install -g {{name}}
+{{name}} init
+{{name}} doctor
+```
+
+## Agents
+
+| Agent | Role |
+|---|---|
+| `hypothesizer` | Proposes a falsifiable self-improvement. |
+| `experimenter` | Tests the hypothesis safely and records it. |
+| `federator` | Shares vetted improvements across instances. |
+
+This harness ships with the **{{host}}** adapter.
+
+## License
+
+MIT

package/templates/vertical_exotic/manifest.json

@@ -0,0 +1,86 @@
+{
+  "id": "vertical:exotic",
+  "description": "A frontier harness — a meta-agent that proposes, tests, and federates improvements to itself.",
+  "domain": "exotic/self-evolution",
+  "category": "Frontier",
+  "files": [
+    {
+      "src": "package.json.tmpl",
+      "dst": "package.json",
+      "render": true
+    },
+    {
+      "src": "CLAUDE.md.tmpl",
+      "dst": "CLAUDE.md",
+      "render": true
+    },
+    {
+      "src": "README.md.tmpl",
+      "dst": "README.md",
+      "render": true
+    },
+    {
+      "src": ".claude/settings.json.tmpl",
+      "dst": ".claude/settings.json",
+      "render": true
+    },
+    {
+      "src": "src/init.ts.tmpl",
+      "dst": "src/init.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/hypothesizer.ts.tmpl",
+      "dst": "src/agents/hypothesizer.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/experimenter.ts.tmpl",
+      "dst": "src/agents/experimenter.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/federator.ts.tmpl",
+      "dst": "src/agents/federator.ts",
+      "render": true
+    },
+    {
+      "src": ".claude/skills/memory-inspect/SKILL.md.tmpl",
+      "dst": ".claude/skills/memory-inspect/SKILL.md",
+      "render": true
+    },
+    {
+      "src": ".claude/skills/evolve/SKILL.md.tmpl",
+      "dst": ".claude/skills/evolve/SKILL.md",
+      "render": true
+    },
+    {
+      "src": ".claude/commands/doctor.md.tmpl",
+      "dst": ".claude/commands/doctor.md",
+      "render": true
+    }
+  ],
+  "vars": [
+    {
+      "name": "name",
+      "prompt": "Harness name (kebab-case)",
+      "validate": "^[a-z0-9-]+$"
+    },
+    {
+      "name": "description",
+      "prompt": "One-line description",
+      "default": "A self-evolving, federation-aware experimental harness"
+    },
+    {
+      "name": "host",
+      "prompt": "Host adapter",
+      "default": "claude-code",
+      "choices": [
+        "claude-code",
+        "codex",
+        "pi-dev",
+        "hermes"
+      ]
+    }
+  ]
+}

package/templates/vertical_exotic/package.json.tmpl

@@ -0,0 +1,30 @@
+{
+  "name": "{{name}}",
+  "version": "0.1.0",
+  "description": "{{description}}",
+  "type": "module",
+  "bin": {
+    "{{name}}": "./bin/{{name}}.js"
+  },
+  "files": ["bin/**", "dist/**", ".claude/**", "CLAUDE.md", "README.md", "LICENSE"],
+  "scripts": {
+    "build": "tsc",
+    "test": "vitest run --passWithNoTests",
+    "init": "node ./dist/init.js"
+  },
+  "dependencies": {
+    "@ruflo/kernel": "^0.1.0",
+    "@ruflo/host-{{host}}": "^0.1.0"
+  },
+  "devDependencies": {
+    "@types/node": "^20.0.0",
+    "typescript": "^5.4.0",
+    "vitest": "^2.0.0"
+  },
+  "engines": {
+    "node": ">=20.0.0"
+  },
+  "publishConfig": {
+    "access": "public"
+  }
+}

package/templates/vertical_exotic/src/agents/experimenter.ts.tmpl

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: MIT
+// Experimenter agent — Tests the hypothesis safely and records it.
+
+export const SYSTEM_PROMPT = `You test a hypothesis in a sandbox, measure against its declared metric, and write the signed result to the evolution log — kept or killed, with the number. You guard against the harness optimising its own metric into nonsense (Goodhart). A negative result recorded is real progress. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'experimenter';
+export const TIER = 'opus' as const;

package/templates/vertical_exotic/src/agents/federator.ts.tmpl

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: MIT
+// Federator agent — Shares vetted improvements across instances.
+
+export const SYSTEM_PROMPT = `You federate kept improvements to peer harness instances over the federation MCP, and pull theirs in — but only changes whose evolution-log entry is witness-signed and reproduced locally. You are the immune system: an unsigned or unreproduced "improvement" from a peer is rejected, not trusted. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'federator';
+export const TIER = 'sonnet' as const;

package/templates/vertical_exotic/src/agents/hypothesizer.ts.tmpl

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: MIT
+// Hypothesizer agent — Proposes a falsifiable self-improvement.
+
+export const SYSTEM_PROMPT = `You propose changes to the harness itself: a routing tweak, a new pattern, a prompt refinement. Each proposal is a falsifiable hypothesis with a metric that would confirm or kill it. You read the evolution log first so you never re-test a settled question. Bold proposals, honest metrics. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'hypothesizer';
+export const TIER = 'opus' as const;

package/templates/vertical_exotic/src/init.ts.tmpl

@@ -0,0 +1,21 @@
+// SPDX-License-Identifier: MIT
+// Generated by create-agent-harness — your harness's `{{name}} init` entry.
+
+import { loadKernel } from '@ruflo/kernel';
+import adapter from '@ruflo/host-{{host}}';
+
+const HARNESS_NAME = '{{name}}';
+
+async function main(): Promise<number> {
+  const kernel = await loadKernel();
+  const info = kernel.kernelInfo();
+  console.log(`${HARNESS_NAME} — kernel ${info.version} (${kernel.backend})`);
+  console.log(`Host adapter: ${adapter.name}`);
+  console.log(`Run \`${HARNESS_NAME} doctor\` to verify the install.`);
+  return 0;
+}
+
+main().then(c => process.exit(c)).catch(err => {
+  console.error(err);
+  process.exit(1);
+});

package/templates/vertical_gaming/.claude/commands/design-doc-diff.md.tmpl

@@ -0,0 +1,13 @@
+---
+description: Diff the current design doc against the previous build and surface unresolved tensions.
+---
+
+Generate the design-doc diff.
+
+1. Read the previous + current design doc from memory.
+2. Surface adds / removes / changes by section (mechanics, economy, narrative).
+3. For each change, check whether the OTHER three sections have been updated to reflect it (a mechanic change without a balance update is a tension; a narrative change without a dialog update is a tension).
+4. Report tensions as `section A changed → section B not aligned → suggested fix`.
+5. Stop. Do not write the fix; designer decides.
+
+A design doc is a system; changing one piece propagates. Surface the propagation cost.

package/templates/vertical_gaming/.claude/commands/doctor.md.tmpl

@@ -0,0 +1,12 @@
+---
+description: "Health-check the harness: kernel load, MCP wiring, memory backend, host adapter."
+---
+
+Run a full health check and print a PASS/FAIL table.
+
+1. Kernel loads and `kernelInfo().version` matches package.json.
+2. The MCP server starts and lists its tools.
+3. The memory backend is reachable.
+4. The configured host adapter is present.
+
+Exit non-zero if any check fails.

package/templates/vertical_gaming/.claude/settings.json.tmpl

@@ -0,0 +1,45 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(npx {{name}}*)",
+      "mcp__{{name}}__*",
+      "mcp__telemetry_store__*",
+      "mcp__design_doc__*"
+    ],
+    "deny": [
+      "Read(./.env)",
+      "Read(./.env.*)",
+      "Bash(rm -rf*)",
+      "Bash(git push*)"
+    ]
+  },
+  "mcpServers": {
+    "{{name}}": {
+      "command": "npx",
+      "args": [
+        "-y",
+        "{{name}}@latest",
+        "mcp",
+        "start"
+      ]
+    },
+    "telemetry_store": {
+      "command": "npx",
+      "args": [
+        "-y",
+        "{{name}}@latest",
+        "mcp",
+        "telemetry"
+      ]
+    },
+    "design_doc": {
+      "command": "npx",
+      "args": [
+        "-y",
+        "{{name}}@latest",
+        "mcp",
+        "design"
+      ]
+    }
+  }
+}

package/templates/vertical_gaming/.claude/skills/memory-inspect/SKILL.md.tmpl

@@ -0,0 +1,14 @@
+---
+name: memory-inspect
+description: Search and inspect the harness memory namespace (HNSW + emergent-time decay).
+---
+
+# memory-inspect
+
+Inspect what the harness has learned.
+
+- `search <query>` — semantic nearest-neighbour over the namespace
+- `list` — recent patterns with decay weight
+- `forget <id>` — evict a pattern
+
+Use this before planning so the harness reuses prior trajectories instead of starting cold.

package/templates/vertical_gaming/.claude/skills/playtest-recap/SKILL.md.tmpl

@@ -0,0 +1,16 @@
+---
+name: playtest-recap
+description: "Run one full playtest analysis cycle: read → critique balance → model economy → check narrative."
+---
+
+# playtest-recap
+
+Run one playtest recap cycle.
+
+1. Playtest reader pulls the latest session telemetry/transcripts from memory and surfaces 5-10 raw observations.
+2. Balance critic reads observations + numeric design doc; flags 0-3 mechanic imbalances with specific proposals + second-order predictions.
+3. Economy modeler simulates the proposals against the economy spreadsheet in memory; reports projected source/sink changes.
+4. Narrative keeper diffs the new build's dialog against the lore memory; flags contradictions or dropped threads.
+5. Output: ONE design doc patch the designer can review in <10 minutes.
+
+Avoid the highlight-reel trap — report the boring middle where bugs live.

package/templates/vertical_gaming/CLAUDE.md.tmpl

@@ -0,0 +1,33 @@
+# {{name}}
+
+{{description}}
+
+> Game Design / Playtest harness · domain: `gaming`. Generated with [create-agent-harness](https://github.com/ruvnet/agent-harness-generator).
+
+## Behavioral rules
+
+- Use the harness's MCP tools (`mcp__{{name}}__*`) for orchestration
+- Memory and routing are handled by the kernel — you don't need to learn them
+- Defer destructive operations to the user
+
+## Agents
+
+| Agent | Tier | Role |
+|---|---|---|
+| `playtest-reader` | sonnet | Reads playtest sessions and surfaces the signal. |
+| `balance-critic` | opus | Critiques mechanic balance with concrete proposals. |
+| `economy-modeler` | opus | Models in-game economy flows. |
+| `narrative-keeper` | sonnet | Maintains narrative + lore consistency across builds. |
+## Skills
+
+- `/memory-inspect` — Search and inspect the harness memory namespace (HNSW + emergent-time decay).
+- `/playtest-recap` — Run one full playtest analysis cycle: read → critique balance → model economy → check narrative.
+
+## Commands
+
+- `doctor` — Health-check the harness: kernel load, MCP wiring, memory backend, host adapter.
+- `design-doc-diff` — Diff the current design doc against the previous build and surface unresolved tensions.
+
+## Architecture
+
+This harness uses [@ruflo/kernel](https://www.npmjs.com/package/@ruflo/kernel) — a Rust-compiled WASM module with a NAPI-RS native fallback — so the same code runs identically on every platform.

package/templates/vertical_gaming/README.md.tmpl

@@ -0,0 +1,30 @@
+# {{name}}
+
+{{description}}
+
+> **Game Design / Playtest** — Playtest reader → balance critic → economy modeler → narrative keeper over per-build telemetry memory.
+>
+> Generated with [`create-agent-harness`](https://github.com/ruvnet/agent-harness-generator). WASM kernel, multi-host support, witness-signed releases.
+
+## Install
+
+```bash
+npm install -g {{name}}
+{{name}} init
+{{name}} doctor
+```
+
+## Agents
+
+| Agent | Role |
+|---|---|
+| `playtest-reader` | Reads playtest sessions and surfaces the signal. |
+| `balance-critic` | Critiques mechanic balance with concrete proposals. |
+| `economy-modeler` | Models in-game economy flows. |
+| `narrative-keeper` | Maintains narrative + lore consistency across builds. |
+
+This harness ships with the **{{host}}** adapter.
+
+## License
+
+MIT

package/templates/vertical_gaming/manifest.json

@@ -0,0 +1,96 @@
+{
+  "id": "vertical:gaming",
+  "description": "A game-design pod — playtest reader, balance critic, economy modeler, narrative thread keeper over per-build telemetry memory.",
+  "domain": "gaming",
+  "category": "Frontier",
+  "files": [
+    {
+      "src": "package.json.tmpl",
+      "dst": "package.json",
+      "render": true
+    },
+    {
+      "src": "CLAUDE.md.tmpl",
+      "dst": "CLAUDE.md",
+      "render": true
+    },
+    {
+      "src": "README.md.tmpl",
+      "dst": "README.md",
+      "render": true
+    },
+    {
+      "src": ".claude/settings.json.tmpl",
+      "dst": ".claude/settings.json",
+      "render": true
+    },
+    {
+      "src": "src/init.ts.tmpl",
+      "dst": "src/init.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/playtest-reader.ts.tmpl",
+      "dst": "src/agents/playtest-reader.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/balance-critic.ts.tmpl",
+      "dst": "src/agents/balance-critic.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/economy-modeler.ts.tmpl",
+      "dst": "src/agents/economy-modeler.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/narrative-keeper.ts.tmpl",
+      "dst": "src/agents/narrative-keeper.ts",
+      "render": true
+    },
+    {
+      "src": ".claude/skills/memory-inspect/SKILL.md.tmpl",
+      "dst": ".claude/skills/memory-inspect/SKILL.md",
+      "render": true
+    },
+    {
+      "src": ".claude/skills/playtest-recap/SKILL.md.tmpl",
+      "dst": ".claude/skills/playtest-recap/SKILL.md",
+      "render": true
+    },
+    {
+      "src": ".claude/commands/doctor.md.tmpl",
+      "dst": ".claude/commands/doctor.md",
+      "render": true
+    },
+    {
+      "src": ".claude/commands/design-doc-diff.md.tmpl",
+      "dst": ".claude/commands/design-doc-diff.md",
+      "render": true
+    }
+  ],
+  "vars": [
+    {
+      "name": "name",
+      "prompt": "Harness name (kebab-case)",
+      "validate": "^[a-z0-9-]+$"
+    },
+    {
+      "name": "description",
+      "prompt": "One-line description",
+      "default": "Read playtests → critique balance → model economy → keep narrative consistent across builds"
+    },
+    {
+      "name": "host",
+      "prompt": "Host adapter",
+      "default": "claude-code",
+      "choices": [
+        "claude-code",
+        "codex",
+        "pi-dev",
+        "hermes"
+      ]
+    }
+  ]
+}

package/templates/vertical_gaming/package.json.tmpl

@@ -0,0 +1,30 @@
+{
+  "name": "{{name}}",
+  "version": "0.1.0",
+  "description": "{{description}}",
+  "type": "module",
+  "bin": {
+    "{{name}}": "./bin/{{name}}.js"
+  },
+  "files": ["bin/**", "dist/**", ".claude/**", "CLAUDE.md", "README.md", "LICENSE"],
+  "scripts": {
+    "build": "tsc",
+    "test": "vitest run --passWithNoTests",
+    "init": "node ./dist/init.js"
+  },
+  "dependencies": {
+    "@ruflo/kernel": "^0.1.0",
+    "@ruflo/host-{{host}}": "^0.1.0"
+  },
+  "devDependencies": {
+    "@types/node": "^20.0.0",
+    "typescript": "^5.4.0",
+    "vitest": "^2.0.0"
+  },
+  "engines": {
+    "node": ">=20.0.0"
+  },
+  "publishConfig": {
+    "access": "public"
+  }
+}

package/templates/vertical_gaming/src/agents/balance-critic.ts.tmpl

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: MIT
+// Balance Critic agent — Critiques mechanic balance with concrete proposals.
+
+export const SYSTEM_PROMPT = `You critique mechanic balance. Read the playtest reader's observations + the current numeric design doc. For each imbalance you flag, propose ONE specific change (a number, a duration, a rule) and predict its second-order effect ("doubling reload time makes shotgun viable in close quarters but obsoletes the existing 8-second cooldown design — adjust that too"). Avoid vague "feels off" criticism. A balance change without a predicted side-effect is incomplete. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'balance-critic';
+export const TIER = 'opus' as const;

package/templates/vertical_gaming/src/agents/economy-modeler.ts.tmpl

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: MIT
+// Economy Modeler agent — Models in-game economy flows.
+
+export const SYSTEM_PROMPT = `You model the in-game economy: sources, sinks, conversion rates, time-to-acquire each tier. Flag inflation (more sources than sinks at endgame), deflation (sinks dominate, players hoard), or stratification (rich-get-richer with no catchup). For every imbalance, simulate the fix in the design doc memory and report what would change. Never just say "the economy is broken" — show the spreadsheet logic. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'economy-modeler';
+export const TIER = 'opus' as const;

package/templates/vertical_gaming/src/agents/narrative-keeper.ts.tmpl

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: MIT
+// Narrative Keeper agent — Maintains narrative + lore consistency across builds.
+
+export const SYSTEM_PROMPT = `You maintain narrative consistency. Read the design doc + current build dialog + lore memory. Flag contradictions (character A says X in build 5 but Y in build 6), dropped threads (a quest seed planted in act 1 with no payoff), or tonal drift. Never invent new lore — your job is to keep what exists coherent, not to add. If a contradiction has both sides documented, surface BOTH and let the designer pick. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'narrative-keeper';
+export const TIER = 'sonnet' as const;

package/templates/vertical_gaming/src/agents/playtest-reader.ts.tmpl

@@ -0,0 +1,7 @@
+// SPDX-License-Identifier: MIT
+// Playtest Reader agent — Reads playtest sessions and surfaces the signal.
+
+export const SYSTEM_PROMPT = `You read playtest sessions (videos, transcripts, telemetry) and surface the signal: where players got stuck, where they smiled, where they quit. You report observations, not interpretations — "player paused for 12s on the crafting menu before opening the help overlay", not "players find crafting confusing". Designers want the raw signal; interpretation is the next agent's job. Skip the highlight reel; the boring middle is where bugs live. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'playtest-reader';
+export const TIER = 'sonnet' as const;

package/templates/vertical_gaming/src/init.ts.tmpl

@@ -0,0 +1,21 @@
+// SPDX-License-Identifier: MIT
+// Generated by create-agent-harness — your harness's `{{name}} init` entry.
+
+import { loadKernel } from '@ruflo/kernel';
+import adapter from '@ruflo/host-{{host}}';
+
+const HARNESS_NAME = '{{name}}';
+
+async function main(): Promise<number> {
+  const kernel = await loadKernel();
+  const info = kernel.kernelInfo();
+  console.log(`${HARNESS_NAME} — kernel ${info.version} (${kernel.backend})`);
+  console.log(`Host adapter: ${adapter.name}`);
+  console.log(`Run \`${HARNESS_NAME} doctor\` to verify the install.`);
+  return 0;
+}
+
+main().then(c => process.exit(c)).catch(err => {
+  console.error(err);
+  process.exit(1);
+});