Playtest

Playtest Agent

Deploy Claude as an AI player to find bugs in SENTINEL. The agent interacts with the game through the Bridge API (localhost:3333) and systematically tests game mechanics.

Prerequisites

The Bridge must be running with Claude as the backend. Start it from the project root:

cd sentinel-bridge && deno task start --backend claude

**IMPORTANT**: Always use Claude as the backend for playtesting. This ensures consistent, high-quality GM responses and proper tool handling.

## How to Run

When the user invokes `/playtest`, follow this protocol:

### Step 1: Check Bridge Status and Backend

```bash
# Check if bridge is running
curl -s http://localhost:3333/health

# Check current backend
curl -s http://localhost:3333/state | jq -r '.sentinel.backend // "not running"'

If the bridge isn't running:

1. Start it in a separate terminal: cd C:\dev\SENTINEL\sentinel-bridge && deno task start --backend claude
2. Wait for "Bridge API listening on http://localhost:3333"
3. Return and continue

If the bridge is running but using a different backend, switch to Claude:

curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "slash", "command": "/backend", "args": ["claude"]}'

Verify the backend switched:

curl -s http://localhost:3333/state | jq -r '.sentinel.backend'
# Should output: claude

Step 2: Read the Test Protocol

Read C:\dev\SENTINEL\.claude\skills\playtest\PROTOCOL.md for the testing checklist.

Step 3: Execute Tests

Run through the test scenarios using curl commands. For each test:

1. Execute the action via the Bridge API
2. Verify the response contains expected fields
3. Check state consistency with campaign_state command
4. Log any anomalies

API Commands Reference

# Check health
curl -s http://localhost:3333/health

# Get full state
curl -s http://localhost:3333/state

# Create new campaign
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "slash", "command": "/new", "args": ["Playtest Session"]}'

# Create character (minimal)
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "slash", "command": "/char", "args": ["quick"]}'

# Get campaign state
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "campaign_state"}'

# Start session (triggers GM)
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "slash", "command": "/start", "args": []}'

# Say something to GM
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "say", "text": "I approach the contact cautiously"}'

# Run any slash command
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "slash", "command": "/status", "args": []}'

# Save campaign
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "save"}'

# Load campaign
curl -s -X POST http://localhost:3333/command \
  -H "Content-Type: application/json" \
  -d '{"cmd": "load", "campaign_id": "playtest-session"}'

Step 4: Generate Bug Report

After running tests, create a bug report summarizing:

• Passed tests: Commands that worked as expected
• Failed tests: Commands that returned errors or unexpected results
• State inconsistencies: Mismatches between expected and actual state
• Crashes: Any unhandled exceptions or process failures
• Edge cases: Boundary conditions that behaved unexpectedly

Test Scenarios

Scenario 1: Happy Path

1. Create campaign
2. Create character
3. Start session
4. Make a few player actions
5. Run /debrief
6. Verify state persistence

Scenario 2: Edge Cases

1. Social energy at 0 - try social actions
2. Hostile faction - try favor request
3. Invalid commands - verify error handling
4. Rapid command spam - check for race conditions

Scenario 3: State Consistency

1. After each major action, compare campaign_state output
2. Verify faction standings update correctly
3. Verify social energy changes are reflected
4. Verify gear/loadout changes persist

Bug Severity Levels

• CRITICAL: Crash, data loss, game-breaking
• HIGH: Incorrect mechanics, state corruption
• MEDIUM: UI/output issues, confusing behavior
• LOW: Minor inconsistencies, polish issues

Example Output

# Playtest Report - 2026-01-24

## Summary
- Tests run: 15
- Passed: 12
- Failed: 3
- Warnings: 2

## Critical Issues
None

## High Priority
- `/favor` command fails when NPC has no personal_standing field (KeyError)

## Medium Priority
- `/jobs accept` returns success but job not added to active_jobs list
- Social energy display shows 75 but campaign_state shows 70

## Low Priority
- Typo in Nexus faction description: "surveilance" -> "surveillance"

## Test Details
[Full test log...]

Tips

• Run tests in a fresh campaign to avoid state pollution
• Check both the command response AND the campaign_state after each action
• Pay attention to error messages - they often reveal edge cases
• If something feels wrong, it probably is - investigate further