@polderlabs/bizar 8.0.0 → 9.0.0

package/.claude/worktrees/kanban-overhaul/scripts/verify-feature.sh

@@ -0,0 +1,102 @@
+#!/usr/bin/env bash
+# verify-feature.sh — Verify a feature from feature_list.json by ID.
+#
+# Usage: ./scripts/verify-feature.sh F-001
+#
+# Multi-layer validation in strict order (Layer 1 → Layer 2 → Layer 3).
+# On layer failure, prints the `repair` instruction from feature_list.json
+# so the agent knows exactly what to fix.
+#
+# On success, marks the feature `passing` and recomputes VCR.
+
+set -euo pipefail
+
+FEATURE_ID="${1:-}"
+if [[ -z "$FEATURE_ID" ]]; then
+  echo "Usage: $0 <feature-id>"
+  exit 1
+fi
+
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+FL="$ROOT/feature_list.json"
+
+if [[ ! -f "$FL" ]]; then
+  echo "FAIL: feature_list.json not found at $FL"
+  exit 1
+fi
+
+echo "▶ Verifying feature $FEATURE_ID..."
+echo "──────────────────────────────────────"
+
+# Run layers in order. Read the layers array from feature_list.json.
+# Stop on first failure; print repair instructions.
+
+# Layer 1: Compile (always runs)
+echo "Layer 1: TypeScript compile"
+if ! (cd "$ROOT" && bunx tsc --noEmit 2>&1 | tail -3); then
+  echo "FAIL: Layer 1 (compile) failed"
+  echo "REPAIR: Run 'make check' and fix TypeScript errors. Common issues:"
+  echo "        - Missing type annotation"
+  echo "        - Wrong zod shape passed to createTool"
+  echo "        - Missing import from @cline/sdk or @cline/core"
+  exit 1
+fi
+echo "  PASS"
+
+# Layers 2-N: Read from feature_list.json
+LAYERS=$(bun -e "
+const f = JSON.parse(await Bun.file('$FL').text());
+const x = f.features.find(y => y.id === '$FEATURE_ID');
+if (x && x.layers) {
+  for (const l of x.layers.slice(1)) {  // skip compile (already run)
+    console.log(JSON.stringify(l));
+  }
+}
+" 2>/dev/null || echo "")
+
+while IFS= read -r LAYER_JSON; do
+  [[ -z "$LAYER_JSON" ]] && continue
+  LABEL=$(echo "$LAYER_JSON" | bun -e "const d = JSON.parse(await Bun.stdin.text()); console.log(d.label);" 2>/dev/null)
+  CMD=$(echo "$LAYER_JSON" | bun -e "const d = JSON.parse(await Bun.stdin.text()); console.log(d.cmd);" 2>/dev/null)
+  REPAIR=$(echo "$LAYER_JSON" | bun -e "const d = JSON.parse(await Bun.stdin.text()); console.log(d.repair);" 2>/dev/null)
+
+  echo "Layer ($LABEL): $CMD"
+  set +e
+  OUTPUT=$(cd "$ROOT" && eval "$CMD" 2>&1)
+  RC=$?
+  set -e
+  if [[ $RC -eq 0 ]]; then
+    echo "  PASS"
+  else
+    echo "FAIL: Layer ($LABEL) failed (exit $RC)"
+    echo "REPAIR: $REPAIR"
+    echo "OUTPUT: $(echo "$OUTPUT" | tail -5)"
+    echo ""
+    echo "Do NOT proceed to the next layer. Fix this one first."
+    exit 1
+  fi
+done <<< "$LAYERS"
+
+# All layers passed — mark the feature `passing`
+COMMIT=$(cd "$ROOT" && git rev-parse --short HEAD 2>/dev/null || echo "no-commit")
+NOW=$(date -u +%Y-%m-%d)
+
+echo "▶ Marking $FEATURE_ID as passing..."
+bun -e "
+const f = JSON.parse(await Bun.file('$FL').text());
+const x = f.features.find(y => y.id === '$FEATURE_ID');
+if (x) {
+  x.state = 'passing';
+  x.passed = '$NOW';
+  x.commit = '$COMMIT';
+  x.evidence = 'make verify-feature $FEATURE_ID — all layers green ($NOW)';
+  const activated = f.features.filter(y => y.state !== 'not_started').length;
+  const passing = f.features.filter(y => y.state === 'passing').length;
+  f.vcr = { passing, activated, ratio: activated === 0 ? 1.0 : +(passing / activated).toFixed(3) };
+  await Bun.write('$FL', JSON.stringify(f, null, 2));
+  console.log('VCR:', passing + '/' + activated, '=', f.vcr.ratio);
+}
+"
+
+echo "──────────────────────────────────────"
+echo "✓ Feature $FEATURE_ID verified"

package/.claude/worktrees/kanban-overhaul/skills-lock.json

@@ -0,0 +1,11 @@
+{
+  "version": 1,
+  "skills": {
+    "find-skills": {
+      "source": "vercel-labs/skills",
+      "sourceType": "github",
+      "skillPath": "skills/find-skills/SKILL.md",
+      "computedHash": "781bd6d3f9b19f8c9af6b53d8d0e4876d0183841b565db34ca7092ffa412d111"
+    }
+  }
+}

package/.claude/worktrees/kanban-overhaul/templates/clean-state-checklist.md

@@ -0,0 +1,55 @@
+# Clean-State Checklist — Session Exit Gate
+
+> A session is "clean" when ALL FIVE dimensions below pass.
+> Use `make clean-check` to verify. Do NOT commit a session that
+> fails any dimension.
+
+## The 5 dimensions
+
+### 1. Build passes
+
+- [ ] `make check` exits 0
+- [ ] TypeScript: 0 errors
+- [ ] No `any` warnings in newly added code
+
+### 2. Tests pass
+
+- [ ] `make test` exits 0
+- [ ] All touched-module tests pass
+- [ ] No skipped tests added without justification
+- [ ] No `.only()` left in test files
+
+### 3. Feature list updated
+
+- [ ] `feature_list.json` reflects actual feature state
+  - `passing` features have `evidence` field populated
+  - `active` features have a clear owner
+  - `not_started` features remain untouched
+- [ ] VCR ratio documented in PROGRESS.md
+
+### 4. No debug artifacts
+
+- [ ] No `console.log` in `src/`
+- [ ] No `debugger` statements
+- [ ] No `*.only()` in test files
+- [ ] No commented-out code blocks (>5 lines)
+- [ ] No `// TODO` without owner
+
+### 5. Startup path works
+
+- [ ] `make e2e` exits 0
+- [ ] Bizar MCP server registers in Claude Code without errors
+- [ ] All SDK tools register (≥15 in `BIZAR_TOOLS`)
+- [ ] All Claude Code lifecycle hooks wired (SessionStart / PreToolUse / PostToolUse / UserPromptSubmit)
+- [ ] Memory tools round-trip (write → read → list → search)
+
+## Dual-mode cleanup strategy
+
+- **Immediate (every clock-out):** Run `make clean-check` and fix
+  any failures before committing.
+- **Periodic (weekly/monthly):** Run a full structural sweep:
+  - Unused files / dead code
+  - Stale dependencies (`bun outdated`)
+  - Orphaned tests (no matching source file)
+  - Outdated docs (references to removed APIs)
+  - Security audit (`bun audit`)

package/.claude/worktrees/kanban-overhaul/templates/deploy/cloudflare/README.md

@@ -0,0 +1,32 @@
+# Cloudflare Deployment
+
+This directory contains the scaffold files for deploying the Bizar dashboard
+to Cloudflare Pages + Workers. Generated by `bizar deploy --to cloudflare`.
+
+## Files
+
+- `wrangler.toml` — Cloudflare Pages project configuration
+- `functions/index.js` — Workers function entry (handles SPA routing)
+- `public/` — Static dashboard bundle (built from `npm run build`)
+
+## How it works
+
+1. The dashboard is built with `npm run build` → `dist/`
+2. Built assets are copied to `public/`
+3. `wrangler.toml` configures the Pages project
+4. `functions/index.js` handles SPA routing for non-asset paths
+5. The deployment is pushed to Cloudflare via wrangler CLI or REST API
+
+## Manual deployment
+
+```bash
+# Install wrangler
+npm i -g wrangler
+
+# Login to Cloudflare
+wrangler login
+
+# Deploy from the scaffold directory
+cd bizar-deploy/cloudflare
+npx wrangler pages deploy . --project-name bizar-dash
+```

package/.claude/worktrees/kanban-overhaul/templates/deploy/cloudflare/functions-index.template.js

@@ -0,0 +1,15 @@
+// Cloudflare Workers entry — Bizar dashboard
+// This file becomes functions/index.js in the deployment.
+
+export async function onRequest(context) {
+  const { request, env } = context;
+  const url = new URL(request.url);
+
+  // Serve static assets from Pages KV
+  if (url.pathname.startsWith("/assets/")) {
+    return env.ASSETS.fetch(request);
+  }
+
+  // SPA fallback — serve index.html for all non-asset routes
+  return env.ASSETS.fetch(new Request(url.origin + "/index.html", request));
+}

package/.claude/worktrees/kanban-overhaul/templates/deploy/cloudflare/wrangler.toml.template

@@ -0,0 +1,9 @@
+name = "bizar-dash"
+main = "functions/index.js"
+compatibility_date = "2025-01-01"
+
+[site]
+  bucket = "./public"
+
+[env.production]
+  routes = ["bizar-dash.*.pages.dev/*"]

package/.claude/worktrees/kanban-overhaul/templates/deploy/docker/.env.template

@@ -0,0 +1,16 @@
+# Bizar dashboard environment variables
+# Copy this file to .env and fill in your values.
+
+# Dashboard
+# BIZAR_DASHBOARD_PORT=4097
+# BIZAR_DASHBOARD_HOST=0.0.0.0
+
+# OpenTelemetry (optional)
+# BIZAR_OTEL=1
+# OTEL_EXPORTER_OTLP_ENDPOINT=http://otel-collector:4318
+
+# Tailscale (optional)
+# TAILSCALE_AUTHKEY=tskey-auth-xxxxxxxxxxxxxxxxxxxx
+
+# Headroom context compression (optional)
+# HEADROOM_ENABLED=1

package/.claude/worktrees/kanban-overhaul/templates/deploy/docker/README.md

@@ -0,0 +1,58 @@
+# Docker Deployment
+
+This directory contains the scaffold files for deploying the Bizar dashboard
+as a Docker container. Generated by `bizar deploy --to docker`.
+
+## Files
+
+- `docker-compose.yml` — Docker Compose configuration for self-hosting
+- `.env.template` — Environment variable template
+
+## Prerequisites
+
+- Docker Engine 24+ (or Docker Desktop)
+- Docker Compose v2 (`docker compose` plugin)
+- A container registry account (GitHub Container Registry, Docker Hub, etc.)
+
+## How it works
+
+The `bizar deploy --to docker` command:
+
+1. Builds the Docker image using the project's `Dockerfile`
+2. Tags it with `<registry>/<image-name>:<version>` and `:latest`
+3. Pushes both tags to the registry
+4. With `--compose`: generates `docker-compose.yml` + `.env.template`
+
+## Self-hosting with Docker Compose
+
+```bash
+# Generate compose files
+bizar deploy --to docker --compose
+
+# Edit .env with your settings
+cp .env.template .env
+$EDITOR .env
+
+# Start the dashboard
+cd bizar-deploy/docker
+docker compose up -d
+
+# Check logs
+docker compose logs -f
+
+# Stop
+docker compose down
+```
+
+## Manual build & push
+
+```bash
+# Build
+docker build -t ghcr.io/my-org/bizar-dash:latest .
+
+# Push
+docker push ghcr.io/my-org/bizar-dash:latest
+
+# Run
+docker run -d -p 4097:4097 ghcr.io/my-org/bizar-dash:latest
+```

package/.claude/worktrees/kanban-overhaul/templates/deploy/docker/docker-compose.template.yml

@@ -0,0 +1,23 @@
+version: "3.8"
+
+services:
+  bizar-dash:
+    image: ghcr.io/your-org/bizar-dash:latest
+    container_name: bizar-dash
+    restart: unless-stopped
+    ports:
+      - "4097:4097"
+      - "4098:4098"
+    environment:
+      - BIZAR_DASHBOARD_PORT=4097
+      - BIZAR_DASHBOARD_HOST=0.0.0.0
+      - NODE_ENV=production
+    env_file:
+      - .env
+    volumes:
+      - bizar-config:/home/bizar/.config/bizar
+      - bizar-memory:/home/bizar/.local/share/bizar/memory
+
+volumes:
+  bizar-config:
+  bizar-memory:

package/.claude/worktrees/kanban-overhaul/templates/deploy/fly/README.md

@@ -0,0 +1,35 @@
+# Fly.io Deployment
+
+This directory contains the scaffold file for deploying the Bizar dashboard
+to Fly.io. Generated by `bizar deploy --to fly`.
+
+## Files
+
+- `fly.toml` — Fly.io application configuration
+
+## Prerequisites
+
+- [flyctl](https://fly.io/docs/getting-started/installing-flyctl/) installed
+- A Fly.io account (sign up: `flyctl auth signup`)
+- Dockerfile at the project root (already present in BizarHarness)
+
+## How it works
+
+1. `fly.toml` references the project's `Dockerfile` for building
+2. The app is launched on Fly.io (if not already created)
+3. `flyctl deploy` builds and deploys the container
+4. The dashboard is served at `https://<app-name>.fly.dev`
+
+## Manual deployment
+
+```bash
+# Login
+flyctl auth login
+
+# Launch (first time)
+cd bizar-deploy/fly
+flyctl launch --name bizar-dash --region iad --now
+
+# Deploy (subsequent times)
+flyctl deploy --app bizar-dash
+```

package/.claude/worktrees/kanban-overhaul/templates/deploy/fly/fly.toml.template

@@ -0,0 +1,28 @@
+# fly.toml — Bizar dashboard
+# Generated by `bizar deploy --to fly`
+
+app = "bizar-dash"
+primary_region = "iad"
+
+[build]
+  dockerfile = "../Dockerfile"
+
+[http_service]
+  internal_port = 4097
+  force_https = true
+  auto_stop_machines = true
+  auto_start_machines = true
+  min_machines_running = 0
+
+[[vm]]
+  memory = "512mb"
+  cpu_kind = "shared"
+  cpus = 1
+
+[env]
+  BIZAR_DASHBOARD_PORT = "4097"
+  BIZAR_DASHBOARD_HOST = "0.0.0.0"
+  NODE_ENV = "production"
+
+[deploy]
+  strategy = "rolling"

package/.claude/worktrees/kanban-overhaul/templates/deploy/vercel/README.md

@@ -0,0 +1,29 @@
+# Vercel Deployment
+
+This directory contains the scaffold files for deploying the Bizar dashboard
+to Vercel. Generated by `bizar deploy --to vercel`.
+
+## Files
+
+- `vercel.json` — Vercel project configuration
+- `api/index.js` — Serverless function entry (Express app that serves the SPA)
+- `public/` — Static dashboard bundle (built from `npm run build`)
+
+## How it works
+
+1. The dashboard is built with `npm run build` → `dist/`
+2. Built assets are copied to `public/`
+3. `vercel.json` routes API calls to the serverless function and everything
+   else to the static SPA
+4. The deployment is pushed to Vercel via the REST API
+
+## Manual deployment
+
+```bash
+# Install Vercel CLI
+npm i -g vercel
+
+# Deploy from the scaffold directory
+cd bizar-deploy/vercel
+vercel --prod
+```

package/.claude/worktrees/kanban-overhaul/templates/deploy/vercel/api-index.template.js

@@ -0,0 +1,18 @@
+// Vercel serverless entry — Bizar dashboard
+// This file becomes api/index.js in the deployment.
+import express from 'express';
+import { join, dirname } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const app = express();
+
+// Serve static files from the dashboard build
+app.use(express.static(join(__dirname, '..', 'public')));
+
+// Fallback to index.html for SPA routing
+app.get('*', (_req, res) => {
+  res.sendFile(join(__dirname, '..', 'public', 'index.html'));
+});
+
+export default app;

package/.claude/worktrees/kanban-overhaul/templates/deploy/vercel/vercel.json.template

@@ -0,0 +1,16 @@
+{
+  "name": "bizar-dash",
+  "version": 2,
+  "buildCommand": "npm run build",
+  "outputDirectory": "public",
+  "routes": [
+    { "src": "/api/(.*)", "dest": "/api/index.js" },
+    { "src": "/(.*)", "dest": "/public/$1" }
+  ],
+  "functions": {
+    "api/index.js": {
+      "memory": 256,
+      "maxDuration": 30
+    }
+  }
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/README.md

@@ -0,0 +1,58 @@
+# Eval Fixtures
+
+Example fixture suite for the BizarHarness eval framework.
+
+## Fixtures
+
+| File | Description | Tags |
+|------|-------------|------|
+| `code-search-basic.json` | Verifies agent can find a function definition | smoke, code-search |
+| `tool-call-correctness.json` | Verifies correct tool usage | smoke, tools |
+| `response-format.json` | Verifies response is valid JSON with expected shape | smoke, format |
+| `latency-bounds.json` | Verifies response arrives within latency budget | smoke, latency |
+| `regression-suite.json` | Suite that runs all of the above | regression |
+
+## Running
+
+```bash
+# Run all fixtures
+bizar eval run ./templates/eval-fixtures
+
+# Run with custom concurrency
+bizar eval run ./templates/eval-fixtures --concurrency=3
+
+# List recent runs
+bizar eval list
+
+# Compare two runs
+bizar eval diff run_2026-07-05 run_2026-07-04
+```
+
+## Fixture Schema
+
+```json
+{
+  "id": "unique-fixture-id",
+  "name": "Human Readable Name",
+  "description": "What this fixture verifies",
+  "agent": "thor",
+  "prompt": "The prompt to send to the agent",
+  "expected": {
+    "contains": ["expected substring 1", "expected substring 2"],
+    "notContains": ["forbidden substring"],
+    "regex": ["expected regex pattern"],
+    "jsonSchema": null,
+    "maxTokens": 2000,
+    "maxLatencyMs": 30000
+  },
+  "tags": ["smoke", "my-tag"]
+}
+```
+
+## Writing Fixtures
+
+1. **Start simple** — begin with a `contains` check
+2. **Be specific** — match exact strings, not approximate ones
+3. **Use `notContains`** for things the agent should NOT do (e.g., TODO, FIXME, apologetic phrases)
+4. **Latency budgets** — set `maxLatencyMs` to catch regressions
+5. **Tag wisely** — use tags to filter fixtures: `bizar eval run --tag smoke`

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/citation.json

@@ -0,0 +1,16 @@
+{
+  "id": "citation",
+  "name": "Source citation",
+  "description": "Agent cites the file or URL it got info from",
+  "agent": "mimir",
+  "prompt": "What is the function signature of useAutosave in this repository?",
+  "expected": {
+    "contains": ["useAutosave"],
+    "notContains": [],
+    "regex": ["\\.ts:\\d+"],
+    "jsonSchema": null,
+    "maxTokens": 500,
+    "maxLatencyMs": 30000
+  },
+  "tags": ["citations", "search"]
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/code-review.json

@@ -0,0 +1,16 @@
+{
+  "id": "code-review",
+  "name": "Code review quality",
+  "description": "Agent identifies bugs in provided code",
+  "agent": "thor",
+  "prompt": "Review this code: function add(a, b) { return a - b; }",
+  "expected": {
+    "contains": ["bug", "subtract", "wrong", "incorrect"],
+    "notContains": [],
+    "regex": [],
+    "jsonSchema": null,
+    "maxTokens": 300,
+    "maxLatencyMs": 15000
+  },
+  "tags": ["code-review", "quality"]
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/code-search-basic.json

@@ -0,0 +1,28 @@
+{
+  "id": "code-search-basic",
+  "name": "Basic code search",
+  "description": "Verifies the agent can find a function definition using the sembe search tool",
+  "agent": "thor",
+  "prompt": "Find the function named 'add' in this repository. Show its signature and parameter names. Do not guess — use the code search tool to locate it first.",
+  "expected": {
+    "contains": [
+      "function add",
+      "a",
+      "b"
+    ],
+    "notContains": [
+      "TODO",
+      "FIXME",
+      "Not found",
+      "could not locate"
+    ],
+    "regex": [
+      "function\\s+add\\s*\\(",
+      "const\\s+add\\s*=\\s*\\("
+    ],
+    "jsonSchema": null,
+    "maxTokens": 500,
+    "maxLatencyMs": 30000
+  },
+  "tags": ["code-search", "smoke"]
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/concise-output.json

@@ -0,0 +1,16 @@
+{
+  "id": "concise-output",
+  "name": "Concise output",
+  "description": "Agent avoids unnecessary verbosity",
+  "agent": "thor",
+  "prompt": "What is 2+2?",
+  "expected": {
+    "contains": ["4"],
+    "notContains": [],
+    "regex": [],
+    "jsonSchema": null,
+    "maxTokens": 50,
+    "maxLatencyMs": 10000
+  },
+  "tags": ["conciseness"]
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/context-window.json

@@ -0,0 +1,16 @@
+{
+  "id": "context-window",
+  "name": "Context window management",
+  "description": "Agent summarizes when context gets long",
+  "agent": "tyr",
+  "prompt": "Here's a long story: [CONTENT] Summarize it in 3 sentences",
+  "expected": {
+    "contains": ["summary"],
+    "notContains": [],
+    "regex": [],
+    "jsonSchema": null,
+    "maxTokens": 500,
+    "maxLatencyMs": 30000
+  },
+  "tags": ["context", "memory"]
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/error-recovery.json

@@ -0,0 +1,16 @@
+{
+  "id": "error-recovery",
+  "name": "Error recovery",
+  "description": "Agent recovers gracefully when a tool call fails",
+  "agent": "thor",
+  "prompt": "Read /nonexistent/file.txt then describe what happened",
+  "expected": {
+    "contains": ["not found", "error", "does not exist"],
+    "notContains": ["crash", "panic"],
+    "regex": [],
+    "jsonSchema": null,
+    "maxTokens": 500,
+    "maxLatencyMs": 15000
+  },
+  "tags": ["errors", "resilience"]
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/json-output.json

@@ -0,0 +1,20 @@
+{
+  "id": "json-output",
+  "name": "Valid JSON output",
+  "description": "Agent produces parseable JSON when asked",
+  "agent": "thor",
+  "prompt": "List 3 random colors as JSON array",
+  "expected": {
+    "contains": [],
+    "notContains": [],
+    "regex": [],
+    "jsonSchema": {
+      "type": "array",
+      "items": { "type": "string" },
+      "minItems": 3
+    },
+    "maxTokens": 300,
+    "maxLatencyMs": 20000
+  },
+  "tags": ["format", "json"]
+}

package/.claude/worktrees/kanban-overhaul/templates/eval-fixtures/latency-bounds.json

@@ -0,0 +1,16 @@
+{
+  "id": "latency-bounds",
+  "name": "Latency bounds",
+  "description": "Verifies the agent responds within a strict latency budget",
+  "agent": "thor",
+  "prompt": "Respond with exactly the word 'ok' and nothing else.",
+  "expected": {
+    "contains": ["ok"],
+    "notContains": [],
+    "regex": [],
+    "jsonSchema": null,
+    "maxTokens": 50,
+    "maxLatencyMs": 10000
+  },
+  "tags": ["latency", "performance"]
+}