Kiln — Agent Skill Definition

You are controlling a physical 3D printer through Kiln. Physical actions are irreversible and can damage hardware. Follow these rules strictly.

Quick Start

kiln setup          # interactive wizard — finds printers, saves config
kiln verify         # check everything is working
kiln status --json  # see what the printer is doing

Then ask the human what they'd like to print.

---

Which Interface to Use

Kiln supports two interfaces. Pick based on your capabilities:

Don't know which? Try CLI first. Run kiln status --json. If that works, you're good. MCP gives you more tools but requires server setup.

---

CLI Interface

Run commands via your shell/exec tool. Always use --json for machine-readable output.

kiln <command> [options] --json

First-Time Setup

If the printer isn't configured yet, run these first:

# Interactive wizard: auto-discovers printers, saves config, tests connection
kiln setup

# Or manually add a printer
kiln auth --name my-printer --host http://192.168.1.100 --type octoprint --api-key YOUR_KEY

# Verify everything works (Python, slicer, config, printer reachable, database)
kiln verify

# Scan network for printers
kiln discover --json

After setup, config is saved to ~/.kiln/config.yaml — no env vars needed.

Core Commands

# Check printer status (start here)
kiln status --json

# List files on printer
kiln files --json

# Run safety checks before printing
kiln preflight --json
kiln preflight --material PLA --json

# Upload a G-code file
kiln upload /path/to/model.gcode --json

# Start printing (auto-uploads local files, auto-runs preflight)
kiln print model.gcode --json
kiln print model.gcode --dry-run --json   # preview without starting

# Cancel / pause / resume
kiln cancel --json
kiln pause --json
kiln resume --json

# Set temperatures
kiln temp --tool 210 --bed 60 --json
kiln temp --json                        # read current temps (no flags)

# Send raw G-code
kiln gcode G28 "G1 X50 Y50 F3000" --json

# Slice STL to G-code
kiln slice model.stl --json
kiln slice model.stl --print-after --json   # slice + upload + print

# Webcam snapshot
kiln snapshot --save photo.jpg --json

# Wait for print to finish (blocks until done)
kiln wait --json

# Print history
kiln history --json
kiln history --status completed --json

# Discover printers on network
kiln discover --json

# Cost estimate
kiln cost model.gcode --json

Outsourced Manufacturing (Fulfillment)

No local printer? Printer busy? Kiln can outsource to manufacturing services (Craftcloud, Sculpteo) with the same CLI.

# List available materials from configured service
kiln order materials --json

# Get a manufacturing quote (uploads model, returns pricing + lead time)
kiln order quote model.stl -m pla_standard --json

# Place the order [confirm — ask human first, shows price]
kiln order place QUOTE_ID --json

# Track order status
kiln order status ORDER_ID --json

# Cancel (if still cancellable)
kiln order cancel ORDER_ID --json

# Compare local printing vs. outsourced cost side-by-side
kiln compare-cost model.gcode --fulfillment-material pla_standard --json

Setup: Set one of these env vars (or add to ~/.kiln/config.yaml):

export KILN_CRAFTCLOUD_API_KEY="your_key"     # Craftcloud (easiest — one key)
# OR
export KILN_SCULPTEO_API_KEY="your_key"       # Sculpteo

Agent workflow: Check local printer → if unavailable/busy → quote fulfillment → present price to human → human approves → place order → return tracking link.

Text & Sketch to 3D Generation

Generate printable 3D models from text descriptions or napkin sketches. Kiln auto-discovers providers from environment variables.

# List available generation providers [safe]
kiln generate list --json

# Generate a model from text [confirm — creates new file]
kiln generate "a small vase with organic curves" --provider gemini --json
kiln generate "phone stand" --provider meshy --style organic --json

# Check generation status (for async providers like Meshy/Tripo3D)
kiln generate status JOB_ID --json

# Download completed result
kiln generate download JOB_ID --json

MCP equivalents:

{"name": "list_generation_providers", "arguments": {}}
{"name": "generate_model", "arguments": {"prompt": "a small vase", "provider": "gemini"}}
{"name": "check_generation_status", "arguments": {"job_id": "gemini-abc123"}}
{"name": "download_generated_model", "arguments": {"job_id": "gemini-abc123"}}

Available providers (set env var to enable):

Gemini Deep Think uses Google's Gemini API to reason about geometry and generate precise OpenSCAD code, which is compiled locally to STL. Supports text descriptions and sketch/napkin-drawing descriptions. Requires OpenSCAD installed locally.

Agent workflow: Ask what the user wants → generate with best available provider → validate mesh → slice → print.

Model Marketplace Search

Search and download existing 3D models from online marketplaces before generating from scratch.

# Search across all connected marketplaces [safe]
kiln search "phone stand" --json

# Search a specific marketplace [safe]
kiln search "vase" --marketplace thingiverse --json

# Get model details [safe]
kiln model-details thingiverse MODEL_ID --json

# Download a model file [confirm — downloads to local disk]
kiln model-download thingiverse MODEL_ID --json

MCP equivalents:

{"name": "search_all_models", "arguments": {"query": "phone stand"}}
{"name": "search_models", "arguments": {"query": "vase", "marketplace": "thingiverse"}}
{"name": "get_model_details", "arguments": {"marketplace": "thingiverse", "model_id": "12345"}}
{"name": "download_model_file", "arguments": {"marketplace": "thingiverse", "model_id": "12345"}}

Supported marketplaces: Thingiverse, MyMiniFactory, Thangs, Cults3D, GrabCAD, Etsy.

Agent workflow: User describes what they want → search marketplaces → present top results → if nothing fits, generate from text instead.

Fleet Management

Manage multiple printers as a fleet with job queuing and smart routing.

# Register a printer in the fleet [guarded]
kiln fleet add --name ender3 --host http://192.168.1.100 --type octoprint --json

# Fleet-wide status [safe]
kiln fleet status --json

# Submit a job to the queue (auto-routes to best available printer)
kiln fleet print model.gcode --json

# View job queue [safe]
kiln fleet queue --json

MCP equivalents:

{"name": "fleet_status", "arguments": {}}
{"name": "register_printer", "arguments": {"name": "ender3", "host": "http://192.168.1.100", "type": "octoprint"}}
{"name": "submit_fleet_job", "arguments": {"filename": "model.gcode"}}
{"name": "list_queue", "arguments": {}}

Webhooks

Register HTTP endpoints to receive real-time notifications.

# Register a webhook [guarded]
kiln webhook add https://example.com/hook --events print_complete,print_failed --json

# List webhooks [safe]
kiln webhook list --json

# Delete a webhook [confirm]
kiln webhook delete WEBHOOK_ID --json

All payloads are signed with HMAC-SHA256 for verification.

Multi-Printer Support

# List saved printers
kiln printers --json

# Target a specific printer (works with any command)
kiln --printer my-ender3 status --json
kiln --printer bambu-x1c print model.gcode --json

Run kiln --help for all commands. kiln <command> --help for options.

CLI Response Format

Success — exit code 0, JSON to stdout:

{"status": "printing", "filename": "model.gcode", "progress": 42.5,
 "temps": {"tool": 210.0, "bed": 60.0}}

Error — non-zero exit code, JSON with "error":

{"error": "Printer is offline"}

Warnings — "warnings" array alongside normal data:

{"status": "ok", "warnings": ["Hotend temperature above typical PLA range"]}

Check exit code first (0 = success), then "warnings" in the JSON.

Example Responses

kiln status --json (printing):

{"status": "success", "data": {"printer": {"status": "printing", "temps": {"tool0": {"actual": 210.0, "target": 210.0}, "bed": {"actual": 60.0, "target": 60.0}}}, "job": {"filename": "model.gcode", "progress": 42.5, "time_left": 3600}}}

kiln print model.gcode --json (started):

{"status": "success", "message": "Print started", "filename": "model.gcode"}

kiln order quote model.stl -m pla_standard --json:

{"status": "success", "quote_id": "q_abc123", "price_usd": 12.50, "lead_time_days": 5, "shipping_options": [{"id": "std", "price_usd": 4.99, "days": 7}]}

---

MCP Interface

If your platform has an MCP client, Kiln exposes 430 tools as an MCP server. Tools are called by name with JSON arguments — your MCP client handles the transport.

Starting the MCP Server

kiln serve

Or configure in Claude Desktop (~/.config/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "kiln": {
      "command": "kiln",
      "args": ["serve"],
      "env": {
        "KILN_PRINTER_HOST": "http://your-printer-ip",
        "KILN_PRINTER_API_KEY": "your_key",
        "KILN_PRINTER_TYPE": "octoprint"
      }
    }
  }
}

MCP Tool Call Format

{"name": "printer_status", "arguments": {}}
{"name": "start_print", "arguments": {"filename": "model.gcode"}}
{"name": "set_temperature", "arguments": {"tool_temp": 210, "bed_temp": 60}}
{"name": "send_gcode", "arguments": {"commands": ["G28", "G1 X50 Y50 F3000"]}}
{"name": "upload_file", "arguments": {"file_path": "/path/to/model.gcode"}}

Parameter names and types are auto-documented by the server — your client shows them. Call get_started() for an onboarding guide.

MCP Response Format

All tools return JSON objects. Same pattern as CLI:

• Success: tool-specific fields
• Error: {"error": "message", "status": "error"}
• Warnings: "warnings" array alongside data

---

Setup

Set these environment variables before using Kiln (CLI or MCP):

export KILN_PRINTER_HOST="http://your-printer-ip"
export KILN_PRINTER_API_KEY="your_api_key"
export KILN_PRINTER_TYPE="octoprint"   # or: moonraker, bambu, prusaconnect, elegoo

Verify connectivity:

kiln status --json

File Intelligence

G-code files on printers often have cryptic names (test5112.gcode, spacer_v3.gcode). Kiln extracts metadata from G-code headers so you can reason about files without relying on filenames.

# Analyze a specific file [safe]
kiln analyze-file benchy.gcode --json

Returns structured metadata:

{
  "material": "PLA",
  "estimated_time_seconds": 6150,
  "tool_temp": 210.0,
  "bed_temp": 60.0,
  "slicer": "PrusaSlicer 2.7.0",
  "layer_height": 0.2,
  "filament_used_mm": 4523.45
}

How to use file intelligence:

1. List files: kiln files --json — each file now includes metadata
2. Check material match before printing: compare material field with loaded material (kiln material show --json)
3. Check estimated time: use estimated_time_seconds to assess duration
4. Validate temperatures: compare tool_temp/bed_temp with safety profile

Example: Choosing an overnight print

"I found 3 files on your printer:
- benchy.gcode — PLA, ~45 min, 210°C/60°C
- phone_stand.gcode — PLA, ~2h 10m, 210°C/60°C
- test5112.gcode — PETG, ~8h 30m, 240°C/80°C

You have PLA loaded. phone_stand.gcode is the best match for overnight
(PLA-compatible, reasonable duration). Want me to start it?"

---

Safety Model

Kiln enforces physical safety — it will hard-block commands that exceed temperature limits, contain dangerous G-code, or fail pre-flight checks. You cannot bypass these.

You enforce operational judgment — deciding when to ask the human for confirmation vs. acting autonomously. This document defines those rules.

Autonomy Tiers

The human configures how much autonomous control you have. Check kiln autonomy show --json to see the current level.

Level 1 Constraints

At Level 1, the human pre-configures safety boundaries:

# ~/.kiln/config.yaml
autonomy:
  level: 1
  constraints:
    max_print_time_seconds: 14400     # 4 hours max
    allowed_materials: ["PLA", "PETG"] # only these materials
    max_tool_temp: 260                 # hotend ceiling
    max_bed_temp: 100                  # bed ceiling
    require_first_layer_check: true    # must monitor first layer before leaving print unattended

Your workflow at Level 1:

1. Analyze the file (kiln files --json or kiln analyze-file FILE --json)
2. Check constraints: material in allowed list? Time under limit? Temps OK?
3. If ALL constraints pass → proceed without asking
4. If ANY constraint fails → ask the human, explain which constraint failed

Example: Level 1 autonomous print

File: phone_stand.gcode
  Material: PLA ✓ (in allowed list)
  Time: 2h 10m ✓ (under 4h limit)
  Tool temp: 210°C ✓ (under 260°C limit)
  Bed temp: 60°C ✓ (under 100°C limit)
→ All constraints passed. Starting print autonomously.

Example: Level 1 blocked print

File: test5112.gcode
  Material: PETG ✓
  Time: 8h 30m ✗ (exceeds 4h limit)
→ Constraint failed. Asking human for permission.

Level 2 Full Trust

The human has explicitly given you permission to operate freely. This is typically set with a statement like:

"Every file on my printer is for generic PLA, will print in under 2 hours, and is safe."

At Level 2, you may start prints, set temperatures, and send G-code without asking — but you MUST still:

• Run preflight checks (Kiln does this automatically)
• Respect Kiln's hard safety limits (temperature caps, blocked G-code)
• Report what you did after the fact
• Monitor the print if camera is available

Env Var Override

export KILN_AUTONOMY_LEVEL=1  # Quick override without editing config

Tool Safety Levels

Every command has a safety level. Follow the expected behavior exactly. Autonomy level modifies confirm-level behavior only. Safe, guarded, and emergency levels are unaffected by autonomy settings.

Command Safety Classifications

Safe (read-only, call freely)

Guarded (low-risk, report what you did)

Confirm (ask human first)

Emergency (ask human unless active danger)

Recommended Workflows

Upload and Print

# 1. Dry-run first [safe — preview what will happen]
kiln print model.gcode --dry-run --json

# 2. Start print [confirm — ask human first]
#    kiln print auto-uploads local files AND runs preflight automatically
#    "Ready to print model.gcode with PLA. Proceed?"
kiln print model.gcode --json

# 3. Monitor progress [safe]
kiln status --json              # check periodically
kiln snapshot --save check.jpg  # visual check after first layer

# 4. Wait for completion [safe]
kiln wait --json

kiln print auto-uploads local files and runs pre-flight checks. Use --skip-preflight to bypass, --dry-run to preview without starting.

Temperature Adjustment

# 1. Check current temps [safe]
kiln temp --json

# 2. Set temps [confirm — tell human: "Setting hotend to 210°C, bed to 60°C for PLA. OK?"]
kiln temp --tool 210 --bed 60 --json
# IF warnings: relay them

Emergency Response

# 1. Detect issue
kiln status --json   # check for ERROR state or temp anomalies

# 2. IF thermal runaway or physical danger:
kiln gcode M112 --json   # emergency stop — may bypass confirmation
# Then immediately tell human: "Emergency stop triggered because: {reason}"

# 3. IF quality issue but no immediate danger:
# Ask human: "Detected potential failure. Cancel print?"
kiln cancel --json   # only after human confirms

Print Monitoring Loop

Preferred: Use start_monitored_print (MCP) or kiln watch (CLI). These tools combine starting the print with automatic first-layer monitoring.

# CLI — start print with first-layer monitoring
kiln watch --printer myprinter --delay 120 --checks 3 --interval 60 --json

# MCP — start_monitored_print tool does the same thing

start_monitored_print / kiln watch:

1. Starts the print
2. Waits 2 minutes for first layer to begin
3. Captures 3 webcam snapshots at 1-minute intervals
4. Returns snapshots + heuristic analysis (brightness, variance, warnings)
5. Auto-pauses if a failure is reported with confidence >= 0.8

Monitoring tiers (adapt to your capabilities):

After first-layer check passes, continue periodic monitoring:

1. Every 5-10 minutes: kiln status --json — check for:
   • Temperature anomalies (sudden drops = heater failure)
   • Progress stalls (same % for >10 min = possible jam)
   • Error states
2. On completion: cool down heaters

Autonomous Overnight Print

The full workflow for safe autonomous printing while the user sleeps:

1. Analyze:  kiln analyze-file FILE --json
             → Get material, estimated time, temps from G-code metadata.

2. Check autonomy:  kiln autonomy show --json
             → Verify Level 1+ and constraints pass.

3. Start with monitoring:  start_monitored_print(file_name="FILE")
             → Starts print + automatic first-layer checks.

4. Review first layer snapshots:
   - Vision agents: inspect the images directly
   - Text agents: check heuristic_pass, brightness, variance
   - If issues: print is already auto-paused. Alert human.
   - If clean: print continues unattended.

5. Periodic telemetry:  kiln status --json  (every 10 min)
             → Watch for temp drift, progress stalls, errors.

6. Completion:  kiln temp --tool 0 --bed 0 --json
             → Cool down. Report result to human in morning.

Key safety net: If require_first_layer_check is enabled in autonomy constraints, the agent MUST use start_monitored_print instead of start_print. The autonomy check will remind you with "require_first_layer_check": true in the response.

Operational Policies

Heater Idle Protection

Never set temperatures on an idle printer unless the human explicitly asks for pre-heating. If you do, remind: "Heaters are on. Remember to turn them off when done."

Kiln includes a heater watchdog that automatically cools down heaters left on without an active print. Default timeout is 30 minutes (KILN_HEATER_TIMEOUT). Set to 0 to disable. The watchdog will NOT intervene while a print is active — it only triggers on idle heaters.

Relay All Warnings

When Kiln returns warnings, always relay them to the human verbatim.

Never Generate G-code

Never write or modify G-code. Use kiln slice for slicing, or use pre-sliced files already on the printer.

Material Awareness

Before printing, check loaded material (kiln material show --json). If mismatched with what the G-code expects, warn the human.

First-Layer Monitoring

If camera available, check the first few minutes of a new print with kiln snapshot. If something looks wrong, ask the human before acting.

What Kiln Enforces (you cannot bypass)

Licensing & Feature Tiers

Kiln uses a tiered licensing model. Most features are free forever.

Revenue tracking: Kiln takes a 2.5% platform fee on revenue from models published through Kiln's marketplace pipeline (configurable via KILN_PLATFORM_FEE_PCT, range 0.0–15.0%). Local printing is always free.

License key: Set via KILN_LICENSE_KEY env var or ~/.kiln/license file. No key = free tier. Keys are prefixed kiln_pro_, kiln_biz_, kiln_ent_.

Configuration

Preferred: kiln setup (interactive wizard, saves to ~/.kiln/config.yaml).

Alternative: env vars (useful for Docker/CI):