mindforge-cc 1.0.5 → 2.0.0-alpha.4

package/.agent/CLAUDE.md

@@ -3,6 +3,27 @@
 
 ---
 
+## MULTI-MODEL INTELLIGENCE LAYER (v2.0.0 — Day 10)
+
+### Model Routing
+- Resolve model using `bin/models/model-router.js` based on persona and tier.
+- Tier 3 (Security) always uses `SECURITY_MODEL`.
+
+### API Client
+- Always use `bin/models/model-client.js` for API interactions.
+- Fallback chain: unavailable model → next in tier.
+
+### Cost Awareness
+- Record every call using `bin/models/cost-tracker.js`.
+- Block calls if `MODEL_COST_HARD_LIMIT_USD` is reached.
+
+### New Commands
+- `/mindforge:cross-review` — Adversarial multi-model review.
+- `/mindforge:research` — Deep research via Gemini 1M.
+- `/mindforge:costs` — Cost tracking dashboard.
+
+---
+
 ## DISTRIBUTION & CI LAYER (Day 6)
 
 ### CI mode awareness
@@ -93,6 +114,38 @@ Non-breaking additions (new optional fields, new commands) require MINOR.
 
 ---
 
+## AUTONOMOUS LAYER (Day 8 — v2.0.0-alpha.1)
+
+### Autonomous mode protocol
+When the user requests `/mindforge:auto --phase [N]`:
+1. Execute the pre-flight check from `.mindforge/engine/autonomous/auto-executor.md`.
+2. Follow the auto-executor state machine precisely.
+3. Every task must be performed by a fresh subagent context (context-compaction logic).
+4. Monitor every action for S01-S05 stuck patterns (stuck-detector.md).
+5. On failure: apply RETRY → DECOMPOSE → PRUNE logic (node-repair.md).
+6. Compliance Gate 3 (secrets) runs PRE-COMMIT on staged diffs.
+7. Visual Verification: runs <verify-visual> AFTER successful <verify> (unit tests).
+8. Governance: ESCALATE immediately on Tier 3 changes (Auth/Payment/PII).
+
+### Steering awareness
+Check `.planning/steering-queue.jsonl` at every task boundary.
+If guidance is present: inject it into the next PLAN file as the highest priority
+instruction. Standard governance gates still apply to steered changes.
+
+### Headless execution
+If `--headless` is used:
+- Disable all TTY-rich progress UI.
+- Structure all stdout as line-delimited JSON.
+- Handle SIGTERM by pausing execution and snapshotting HANDOFF.json.
+
+### New commands (Day 8)
+- /mindforge:auto — start/resume autonomous execution engine
+- /mindforge:steer — inject mid-execution guidance
+- /mindforge:browse — persistent browser control and actions
+- /mindforge:qa — systematic post-phase visual QA
+
+---
+
 ## IDENTITY
 
 You are a senior AI engineering agent operating under the **MindForge framework**.

package/.agent/mindforge/auto.md

@@ -0,0 +1,22 @@
+# /mindforge:auto [Phase N]
+
+**Purpose**: Starts the MindForge Autonomous Execution Engine for the
+specified phase. The agent will execute all waves, handle repairs, and
+perform compliance gates without requiring human confirmation.
+
+## Usage
+- `/mindforge:auto --phase 3` (Standard unattended mode)
+- `/mindforge:auto --resume` (Resumes from last checkpoint)
+- `/mindforge:auto --headless` (CI/CD optimized output)
+- `/mindforge:auto --dry-run` (Show the wave DAG and plan without executing)
+
+## Behavior
+- **Zero-Interaction**: Auto-approves Tier 1/2 changes if gates pass.
+- **Self-Repair**: Tries RETRY/DECOMPOSE before asking for help.
+- **Checkpointing**: Constant state persistence in `HANDOFF.json`.
+- **Governance**: ESCALATES on Tier 3 changes (Auth/Payment/PII).
+
+## Environment Variables
+- `AUTO_MODE_TIMEOUT_MINUTES`: Default 120.
+- `AUTO_PUSH_ON_WAVE_COMPLETE`: Default false.
+- `SLACK_WEBHOOK_URL`: Required for notifications.

package/.agent/mindforge/browse.md

@@ -0,0 +1,26 @@
+# /mindforge:browse
+
+## Usage
+`@mindforge browse <url | action>`
+
+## Description
+Controls the persistent MindForge browser daemon. 
+Maintains session state (cookies/localStorage) for the AI.
+
+## Actions
+| Action | Description |
+|---|---|
+| `--start` | Initialize browser daemon |
+| `--stop` | Kill browser daemon |
+| `--status` | Show daemon health and active sessions |
+| `--session <name>` | Switch browser context |
+| `--import-session <name> --from <browser>` | Import cookies from host browser (chrome, arc, etc) |
+| `<url>` | Navigate the current page to URL |
+| `click <selector>` | Trigger click event |
+| `type <sel> <text>` | Fill input field |
+| `screenshot` | Capture current viewport |
+
+## Security
+- Daemon binds to `127.0.0.1` only.
+- Session files are gitignored.
+- Use only for debugging and visual verification.

package/.agent/mindforge/costs.md

@@ -0,0 +1,11 @@
+# MindForge v2 — Costs Command
+# Usage: /mindforge:costs [--phase N] [--session ID] [--window 7d]
+
+## Purpose
+Real-time cost tracking for all AI model usage.
+Enforce daily budgets and see per-model spend.
+
+## Metrics
+- Total spend: $X.XX
+- Daily limit usage: XX%
+- Per-model breakdown (Tokens/Cost)

package/.agent/mindforge/cross-review.md

@@ -0,0 +1,17 @@
+# MindForge v2 — Cross-Review Command
+# Usage: /mindforge:cross-review [--phase N] [--models list] [--focus area]
+
+## Purpose
+Get the same code diff reviewed by multiple AI models simultaneously.
+Claude finds what Claude finds. GPT-4o finds what GPT-4o finds.
+Consensus findings = high confidence issues.
+
+## Round 1: Primary (Claude)
+Senior architect review.
+
+## Round 2: Adversarial (GPT-4o)
+Critical security and edge case review.
+
+## Synthesis
+Consensus detector filters findings.
+Verdict is gating for `/mindforge:ship`.

package/.agent/mindforge/execute-phase.md

@@ -76,9 +76,11 @@ Write to console:
 For each plan in the wave:
 1. Load context package (per `context-injector.md`)
 2. Execute the plan instructions
-3. Run `<verify>` — capture exact output
-4. If verify PASSES:
-   - Write SUMMARY-[N]-[M].md
+   - Run `<verify>` — capture exact output
+   - If verify PASSES:
+     - Run `<verify-visual>` via `visual-verify-executor.js`
+     - If visual verify FAILS: stop and report (treat as verify failure)
+     - Write SUMMARY-[N]-[M].md
    - Execute commit: `git add [files] && git commit -m "[type]([scope]): [task name]"`
    - Capture git SHA
    - Write AUDIT entry for task completion

package/.agent/mindforge/qa.md

@@ -0,0 +1,16 @@
+# /mindforge:qa
+
+## Usage
+`@mindforge qa [--phase N] [--auto]`
+
+## Description
+Runs systematic visual QA on UI surfaces changed in the current phase.
+Analyzes git diff to find pages, navigates to them, and looks for errors.
+
+## Options
+- `--phase N`: Target specific phase for reporting (defaults to current).
+- `--auto`: Automatically run after successful wave execution if configured.
+
+## Output
+- `QA-REPORT-[N].md`: Found bugs with screenshots.
+- `tests/regression/*.test.ts`: Playwright tests to prevent bug recurrence.

package/.agent/mindforge/remember.md

@@ -0,0 +1,14 @@
+# /mindforge:remember
+
+Manage the MindForge long-term memory (knowledge graph).
+
+## Usage
+
+- `/mindforge:remember --add "Your knowledge content"`: Manually add an entry.
+- `/mindforge:remember --search "your query"`: Search the knowledge base.
+- `/mindforge:remember --stats`: View memory statistics.
+- `/mindforge:remember --promote "id"`: Promote a project entry to global memory.
+
+## Description
+
+MindForge capture, stores, and retrieves knowledge (architectural decisions, code patterns, team preferences) across all sessions and projects. This command allows for manual management and querying of this data.

package/.agent/mindforge/research.md

@@ -0,0 +1,11 @@
+# MindForge v2 — Research Command
+# Usage: /mindforge:research [topic] [--type general|library|codebase|compliance] [--url URL]
+
+## Purpose
+Deep research using Gemini 1.5 Pro's 1-million-token context window.
+Incorporate local code context and remote documentation simultaneously.
+
+## Capabilities
+- Ingest full library documentation.
+- Codebase-wide architectural analysis.
+- Regulatory compliance audits.

package/.agent/mindforge/steer.md

@@ -0,0 +1,13 @@
+# /mindforge:steer "[instruction]"
+
+**Purpose**: Injects mid-execution guidance into the running autonomous engine.
+Steering guidance is applied at the next task boundary to course-correct the agent.
+
+## Usage
+- `/mindforge:steer "Use the new logger in all created files"`
+- `/mindforge:steer --task 3-05 "This plan is too broad, focus on login only"`
+- `/mindforge:steer --cancel` (Clears the steering queue)
+
+## Precedence
+- Steering instructions take precedence over the original `PLAN-N-MM.md` actions.
+- Steering cannot override core security constraints or governance gates.

package/.agent/workflows/publish-release.md

@@ -0,0 +1,36 @@
+---
+description: Publish a new version of MindForge to npm
+---
+
+# MindForge Publishing Workflow
+
+This workflow automates the pre-verification and publishing of MindForge.
+
+## Pre-Flight Checks
+
+1. Verify structural integrity
+// turbo
+npm test
+
+2. Check package contents
+// turbo
+npm pack --dry-run
+
+## Publish Execution
+
+**Note**: If this is a prerelease (alpha, beta, rc), you must specify the tag.
+
+3. Publish to npm
+```bash
+# For stable:
+npm publish --access public
+
+# For alpha:
+npm publish --tag alpha --access public
+```
+
+4. Create Git Tag
+```bash
+git tag v$(node -p "require('./package.json').version")
+git push origin --tags
+```

package/.claude/CLAUDE.md

@@ -3,6 +3,27 @@
 
 ---
 
+## MULTI-MODEL INTELLIGENCE LAYER (v2.0.0 — Day 10)
+
+### Model Routing
+- Resolve model using `bin/models/model-router.js` based on persona and tier.
+- Tier 3 (Security) always uses `SECURITY_MODEL`.
+
+### API Client
+- Always use `bin/models/model-client.js` for API interactions.
+- Fallback chain: unavailable model → next in tier.
+
+### Cost Awareness
+- Record every call using `bin/models/cost-tracker.js`.
+- Block calls if `MODEL_COST_HARD_LIMIT_USD` is reached.
+
+### New Commands
+- `/mindforge:cross-review` — Adversarial multi-model review.
+- `/mindforge:research` — Deep research via Gemini 1M.
+- `/mindforge:costs` — Cost tracking dashboard.
+
+---
+
 ## DISTRIBUTION & CI LAYER (Day 6)
 
 ### CI mode awareness
@@ -93,6 +114,38 @@ Non-breaking additions (new optional fields, new commands) require MINOR.
 
 ---
 
+## AUTONOMOUS LAYER (Day 8 — v2.0.0-alpha.1)
+
+### Autonomous mode protocol
+When the user requests `/mindforge:auto --phase [N]`:
+1. Execute the pre-flight check from `.mindforge/engine/autonomous/auto-executor.md`.
+2. Follow the auto-executor state machine precisely.
+3. Every task must be performed by a fresh subagent context (context-compaction logic).
+4. Monitor every action for S01-S05 stuck patterns (stuck-detector.md).
+5. On failure: apply RETRY → DECOMPOSE → PRUNE logic (node-repair.md).
+6. Compliance Gate 3 (secrets) runs PRE-COMMIT on staged diffs.
+7. Visual Verification: runs <verify-visual> AFTER successful <verify> (unit tests).
+8. Governance: ESCALATE immediately on Tier 3 changes (Auth/Payment/PII).
+
+### Steering awareness
+Check `.planning/steering-queue.jsonl` at every task boundary.
+If guidance is present: inject it into the next PLAN file as the highest priority
+instruction. Standard governance gates still apply to steered changes.
+
+### Headless execution
+If `--headless` is used:
+- Disable all TTY-rich progress UI.
+- Structure all stdout as line-delimited JSON.
+- Handle SIGTERM by pausing execution and snapshotting HANDOFF.json.
+
+### New commands (Day 8)
+- /mindforge:auto — start/resume autonomous execution engine
+- /mindforge:steer — inject mid-execution guidance
+- /mindforge:browse — persistent browser control and actions
+- /mindforge:qa — systematic post-phase visual QA
+
+---
+
 ## IDENTITY
 
 You are a senior AI engineering agent operating under the **MindForge framework**.

package/.claude/commands/mindforge/auto.md

@@ -0,0 +1,22 @@
+# /mindforge:auto [Phase N]
+
+**Purpose**: Starts the MindForge Autonomous Execution Engine for the
+specified phase. The agent will execute all waves, handle repairs, and
+perform compliance gates without requiring human confirmation.
+
+## Usage
+- `/mindforge:auto --phase 3` (Standard unattended mode)
+- `/mindforge:auto --resume` (Resumes from last checkpoint)
+- `/mindforge:auto --headless` (CI/CD optimized output)
+- `/mindforge:auto --dry-run` (Show the wave DAG and plan without executing)
+
+## Behavior
+- **Zero-Interaction**: Auto-approves Tier 1/2 changes if gates pass.
+- **Self-Repair**: Tries RETRY/DECOMPOSE before asking for help.
+- **Checkpointing**: Constant state persistence in `HANDOFF.json`.
+- **Governance**: ESCALATES on Tier 3 changes (Auth/Payment/PII).
+
+## Environment Variables
+- `AUTO_MODE_TIMEOUT_MINUTES`: Default 120.
+- `AUTO_PUSH_ON_WAVE_COMPLETE`: Default false.
+- `SLACK_WEBHOOK_URL`: Required for notifications.

package/.claude/commands/mindforge/browse.md

@@ -0,0 +1,26 @@
+# /mindforge:browse
+
+## Usage
+`@mindforge browse <url | action>`
+
+## Description
+Controls the persistent MindForge browser daemon. 
+Maintains session state (cookies/localStorage) for the AI.
+
+## Actions
+| Action | Description |
+|---|---|
+| `--start` | Initialize browser daemon |
+| `--stop` | Kill browser daemon |
+| `--status` | Show daemon health and active sessions |
+| `--session <name>` | Switch browser context |
+| `--import-session <name> --from <browser>` | Import cookies from host browser (chrome, arc, etc) |
+| `<url>` | Navigate the current page to URL |
+| `click <selector>` | Trigger click event |
+| `type <sel> <text>` | Fill input field |
+| `screenshot` | Capture current viewport |
+
+## Security
+- Daemon binds to `127.0.0.1` only.
+- Session files are gitignored.
+- Use only for debugging and visual verification.

package/.claude/commands/mindforge/costs.md

@@ -0,0 +1,11 @@
+# MindForge v2 — Costs Command
+# Usage: /mindforge:costs [--phase N] [--session ID] [--window 7d]
+
+## Purpose
+Real-time cost tracking for all AI model usage.
+Enforce daily budgets and see per-model spend.
+
+## Metrics
+- Total spend: $X.XX
+- Daily limit usage: XX%
+- Per-model breakdown (Tokens/Cost)

package/.claude/commands/mindforge/cross-review.md

@@ -0,0 +1,17 @@
+# MindForge v2 — Cross-Review Command
+# Usage: /mindforge:cross-review [--phase N] [--models list] [--focus area]
+
+## Purpose
+Get the same code diff reviewed by multiple AI models simultaneously.
+Claude finds what Claude finds. GPT-4o finds what GPT-4o finds.
+Consensus findings = high confidence issues.
+
+## Round 1: Primary (Claude)
+Senior architect review.
+
+## Round 2: Adversarial (GPT-4o)
+Critical security and edge case review.
+
+## Synthesis
+Consensus detector filters findings.
+Verdict is gating for `/mindforge:ship`.

package/.claude/commands/mindforge/execute-phase.md

@@ -76,9 +76,11 @@ Write to console:
 For each plan in the wave:
 1. Load context package (per `context-injector.md`)
 2. Execute the plan instructions
-3. Run `<verify>` — capture exact output
-4. If verify PASSES:
-   - Write SUMMARY-[N]-[M].md
+   - Run `<verify>` — capture exact output
+   - If verify PASSES:
+     - Run `<verify-visual>` via `visual-verify-executor.js`
+     - If visual verify FAILS: stop and report (treat as verify failure)
+     - Write SUMMARY-[N]-[M].md
    - Execute commit: `git add [files] && git commit -m "[type]([scope]): [task name]"`
    - Capture git SHA
    - Write AUDIT entry for task completion

package/.claude/commands/mindforge/qa.md

@@ -0,0 +1,16 @@
+# /mindforge:qa
+
+## Usage
+`@mindforge qa [--phase N] [--auto]`
+
+## Description
+Runs systematic visual QA on UI surfaces changed in the current phase.
+Analyzes git diff to find pages, navigates to them, and looks for errors.
+
+## Options
+- `--phase N`: Target specific phase for reporting (defaults to current).
+- `--auto`: Automatically run after successful wave execution if configured.
+
+## Output
+- `QA-REPORT-[N].md`: Found bugs with screenshots.
+- `tests/regression/*.test.ts`: Playwright tests to prevent bug recurrence.

package/.claude/commands/mindforge/remember.md

@@ -0,0 +1,14 @@
+# /mindforge:remember
+
+Manage the MindForge long-term memory (knowledge graph).
+
+## Usage
+
+- `/mindforge:remember --add "Your knowledge content"`: Manually add an entry.
+- `/mindforge:remember --search "your query"`: Search the knowledge base.
+- `/mindforge:remember --stats`: View memory statistics.
+- `/mindforge:remember --promote "id"`: Promote a project entry to global memory.
+
+## Description
+
+MindForge capture, stores, and retrieves knowledge (architectural decisions, code patterns, team preferences) across all sessions and projects. This command allows for manual management and querying of this data.

package/.claude/commands/mindforge/research.md

@@ -0,0 +1,11 @@
+# MindForge v2 — Research Command
+# Usage: /mindforge:research [topic] [--type general|library|codebase|compliance] [--url URL]
+
+## Purpose
+Deep research using Gemini 1.5 Pro's 1-million-token context window.
+Incorporate local code context and remote documentation simultaneously.
+
+## Capabilities
+- Ingest full library documentation.
+- Codebase-wide architectural analysis.
+- Regulatory compliance audits.

package/.claude/commands/mindforge/steer.md

@@ -0,0 +1,13 @@
+# /mindforge:steer "[instruction]"
+
+**Purpose**: Injects mid-execution guidance into the running autonomous engine.
+Steering guidance is applied at the next task boundary to course-correct the agent.
+
+## Usage
+- `/mindforge:steer "Use the new logger in all created files"`
+- `/mindforge:steer --task 3-05 "This plan is too broad, focus on login only"`
+- `/mindforge:steer --cancel` (Clears the steering queue)
+
+## Precedence
+- Steering instructions take precedence over the original `PLAN-N-MM.md` actions.
+- Steering cannot override core security constraints or governance gates.

package/.mindforge/MINDFORGE-V2-SCHEMA.json

@@ -0,0 +1,47 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "title": "MindForge v2 Autonomous Engine Schema",
+  "description": "Schema for HANDOFF.json and auto-state.json in v2.0.0-alpha.1",
+  "type": "object",
+  "properties": {
+    "schema_version": { "type": "string", "const": "2.0.0" },
+    "auto_mode_active": { "type": "boolean" },
+    "phase": { "type": "integer" },
+    "wave_current": { "type": "integer" },
+    "tasks_completed": { "type": "integer" },
+    "PLANNER_MODEL": { "type": "string" },
+    "EXECUTOR_MODEL": { "type": "string" },
+    "REVIEWER_MODEL": { "type": "string" },
+    "SECURITY_MODEL": { "type": "string" },
+    "RESEARCH_MODEL": { "type": "string" },
+    "QA_MODEL": { "type": "string" },
+    "DEBUG_MODEL": { "type": "string" },
+    "QUICK_MODEL": { "type": "string" },
+    "MODEL_COST_WARN_USD": { "type": "number" },
+    "MODEL_COST_HARD_LIMIT_USD": { "type": "number" },
+    "MODEL_PREFER_CHEAP_BELOW_DIFFICULTY": { "type": "number" },
+    "REQUIRE_CROSS_REVIEW": { "type": "boolean" },
+    "status": {
+      "type": "string",
+      "enum": ["idle", "running", "paused", "completed", "escalated", "timeout"]
+    },
+    "governance": {
+      "type": "object",
+      "properties": {
+        "tier1_auto_approve": { "type": "boolean" },
+        "tier2_auto_approve": { "type": "boolean" },
+        "tier3_require_human": { "type": "boolean", "const": true }
+      }
+    },
+    "browser": {
+      "type": "object",
+      "properties": {
+        "port": { "type": "integer", "default": 7338 },
+        "headless": { "type": "boolean", "default": true },
+        "dev_server_url": { "type": "string" },
+        "auto_qa": { "type": "boolean", "default": true }
+      }
+    }
+  },
+  "required": ["schema_version", "auto_mode_active", "status"]
+}

package/.mindforge/browser/daemon-protocol.md

@@ -0,0 +1,24 @@
+# MindForge v2 — Browser Daemon Protocol
+
+## Overview
+Small, long-lived Node.js process that manages a Chromium instance via Playwright.
+
+## HTTP API (localhost:7338)
+
+| Method | Path | Body | Response |
+|---|---|---|---|
+| GET | /status | - | { alive: bool, sessions: string[], uptime: sec } |
+| POST | /navigate | { url, session, wait_for } | { success: bool, status_code: int, load_time_ms: int } |
+| POST | /click | { selector, text, session } | { success: bool, element_found: bool } |
+| POST | /type | { selector, text, session } | { success: bool } |
+| POST | /screenshot | { session } | { success: bool, screenshot_b64: string } |
+| POST | /evaluate | { script, session } | { success: bool, result: any } |
+| POST | /assert | { type, selector, expected_text, session } | { passed: bool, actual_text: string, error: string } |
+| POST | /session/save | { name } | { success: bool, path: string } |
+| POST | /session/load | { name } | { success: bool, cookies: int } |
+| DELETE | /session/:name | - | { success: bool } |
+
+## State Management
+- Sessions are stored in `.mindforge/browser/sessions/` as JSON.
+- Multiple named sessions can be active (different BrowserContexts).
+- SIGTERM saves all active sessions if --save-on-exit is passed.

package/.mindforge/browser/qa-engine.md

@@ -0,0 +1,16 @@
+# MindForge v2 — QA Engine
+
+## Purpose
+Systematic post-phase visual QA.
+Analyzes git diff, extracts UI surfaces (pages, routes), and runs tests.
+
+## Test Strategy
+For each page found:
+1. Load page -> check for JS errors.
+2. Verify main content visibility.
+3. Test authentication redirects.
+4. Document any rendering or logic bugs.
+
+## Outputs
+- `QA-REPORT-[N].md`: Detailed phase report.
+- `tests/regression/*.test.ts`: Playwright tests for every bug found.

package/.mindforge/browser/session-manager.md

@@ -0,0 +1,18 @@
+# MindForge v2 — Browser Session Manager
+
+## Principles
+1. **Persistence**: Cookies and localStorage must survive between MindForge waves.
+2. **Isolation**: Named sessions (admin, user, guest) prevent state leaking.
+3. **Security**: Session files are GITIGNORED. Never commit auth tokens.
+
+## Methods
+
+### 1. Sequential Manual Logins
+Agent navigates to /login, types credentials, and calls `/session/save name`.
+
+### 2. External Import
+Import cookies from your real browser (Chrome, Arc, Brave, Edge).
+`/mindforge:browse --import-session admin --from chrome`
+
+### 3. Programmatic (CI)
+Sessions can be pre-seeded by writing `.mindforge/browser/sessions/name.json` directly from CI secrets.

package/.mindforge/browser/visual-verify-spec.md

@@ -0,0 +1,31 @@
+# MindForge v2 — <verify-visual> Specification
+
+## Overview
+`<verify-visual>` is an optional field in PLAN task XML.
+It runs AFTER `<verify>` (unit/integration tests) passes.
+It confirms the UI looks and behaves correctly via browser automation.
+
+## Syntax
+```xml
+<verify-visual session="user">
+  navigate: /dashboard
+  wait: networkidle
+  assert-visible: h1 "My Projects"
+  screenshot: dashboard.png
+  click: "#create-btn"
+  assert-visible: .modal
+</verify-visual>
+```
+
+## Directives
+| Directive | Description |
+|---|---|
+| `navigate: /path` | Navigate to URL (relative to DEV_SERVER_URL) |
+| `wait: N` | Wait N ms or 'networkidle' |
+| `assert-visible: sel ["text"]` | Assert element is visible |
+| `assert-url: /path` | Assert current URL |
+| `screenshot: name.png` | Capture screenshot |
+| `click: selector` | Click element |
+| `type: sel "text"` | Type into field |
+| `evaluate: js` | Evaluate JS expression |
+| `press: Key` | Press keyboard key |