# Deep Doppelganger Implementation Plan > **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking. **Goal:** Build the live `deep_doppelganger` TouchDesigner project in the running TD session. **Architecture:** Use existing tdmcp-agent builders for MediaPipe tracking and major visual systems, then run one controlled TD Python integration pass to create the final compositor, controls, and `out_video`. Keep the project arranged in clear zones and validate through the live bridge. **Tech Stack:** TouchDesigner 2025.32820, tdmcp bridge 0.6.1, tdmcp-agent, MediaPipe TouchDesigner package, TD Python via `exec python`. --- ## Task 1: Runtime Preflight **Files:** - Read: `docs/superpowers/specs/2026-06-08-deep-doppelganger-design.md` - Live TD: `/project1` - [ ] **Step 1: Confirm bridge health** Run: ```bash rtk node dist/cli/agent.js info --output json --no-color ``` Expected: `connected: true`, endpoint `http://127.0.0.1:9980`, `bridge_stale: false`. - [ ] **Step 2: Confirm MediaPipe package path** Run: ```bash rtk proxy sh -lc 'find "$HOME/.tdmcp/packages" "$HOME/tdmcp-packages" -maxdepth 5 -iname "MediaPipe.tox" 2>/dev/null | head -20' ``` Expected: at least `/Users/pantani/.tdmcp/packages/installed/mediapipe-touchdesigner/release/toxes/MediaPipe.tox`. ## Task 2: Build Tracking Sources **Files:** - Live TD: `/project1/MediaPipe`, `/project1/mp_hand_adapter`, body tracking outputs - [ ] **Step 1: Build body tracking** Run: ```bash rtk node dist/cli/agent.js body-tracking --params '{"parent_path":"/project1","build_skeleton":true}' --output json --no-color ``` Expected: MediaPipe engine is loaded, a body tracking adapter/skeleton is created, and the output includes the created paths. - [ ] **Step 2: Build hand tracking** Run: ```bash rtk node dist/cli/agent.js hand-tracking --params '{"parent_path":"/project1","max_hands":2,"coordinate_space":"world","adapter_name":"mp_hand_adapter"}' --output json --no-color ``` Expected: hand landmark CHOP output exists and can be read by the integration pass. ## Task 3: Build Visual Entities **Files:** - Live TD: `/project1/deep_doppelganger` - [ ] **Step 1: Create the root container** Run: ```bash rtk node dist/cli/agent.js container --params '{"parent_path":"/project1","name":"deep_doppelganger"}' --output json --no-color ``` Expected: `/project1/deep_doppelganger` exists. - [ ] **Step 2: Build optical flow from the camera** Run: ```bash rtk node dist/cli/agent.js optical-flow --params '{"name":"motion_flow","parent_path":"/project1/deep_doppelganger","source":"/project1/MediaPipe/video","resolution":[960,540],"sensitivity":5.5,"smoothing":0.72,"blur":2,"direction_from":"edges"}' --output json --no-color ``` Expected: a motion-flow container exists and returns an output TOP path. - [ ] **Step 3: Build spectral point cloud** Run: ```bash rtk node dist/cli/agent.js point-cloud --params '{"source":"existing","existing":"/project1/MediaPipe/video","resolution":192,"depth_scale":1.65,"point_size":0.018,"rotate":8,"expose_controls":true,"parent_path":"/project1/deep_doppelganger"}' --output json --no-color ``` Expected: a point-cloud container exists and returns an output TOP path. - [ ] **Step 4: Build feedback tunnel from the camera** Run: ```bash rtk node dist/cli/agent.js feedback-tunnel --params '{"name":"feedback_tunnel","parent_path":"/project1/deep_doppelganger","source":"/project1/MediaPipe/video","zoom":1.035,"rotate":1.6,"hue_shift":0.012,"decay":0.93,"resolution":[1280,720]}' --output json --no-color ``` Expected: a feedback tunnel container exists and returns an output TOP path. ## Task 4: Integrate Final Composite And Controls **Files:** - Create: `_workspace/deep_doppelganger/integrate_deep_doppelganger.py` - Live TD: `/project1/deep_doppelganger/out_video` - [ ] **Step 1: Create the integration script** Write `_workspace/deep_doppelganger/integrate_deep_doppelganger.py` with TD Python that: - Creates or reuses `/project1/deep_doppelganger`. - Creates select TOPs for camera, motion flow, point cloud, and feedback tunnel. - Builds `ghost_delay`, `flow_glitch`, `layer_mixer`, and `out_video`. - Adds custom controls: `Shock`, `Ghostdelay`, `Depthscale`, `Tunnelopen`, `Glitchbite`, `Trackingpanic`, `Cameraopacity`, `Blackout`. - Arranges nodes in non-overlapping zones. - Prints a JSON report with `out_video`, created nodes, warnings, and missing sources. - [ ] **Step 2: Generate the exec payload** Create the JSON parameters file consumed by `exec python`: ```bash rtk node -e 'const fs=require("fs"); const script=fs.readFileSync("_workspace/deep_doppelganger/integrate_deep_doppelganger.py","utf8"); fs.writeFileSync("_workspace/deep_doppelganger/integrate_payload.json", JSON.stringify({script, return_output:true}, null, 2));' ``` Expected: `_workspace/deep_doppelganger/integrate_payload.json` exists and contains a non-empty `script` field plus `return_output: true`. - [ ] **Step 3: Execute the integration script** Run: ```bash rtk node dist/cli/agent.js exec python --allow-unsafe --params-file _workspace/deep_doppelganger/integrate_payload.json --output json --no-color ``` Expected: `/project1/deep_doppelganger/out_video` exists at `1280x720` in this Non-Commercial TD session and the report says `ok: true`. ## Task 5: Output And Validation **Files:** - Live TD: `/project1/deep_doppelganger/out_video` - [ ] **Step 1: Create output window** Run: ```bash rtk node dist/cli/agent.js output --params '{"source_path":"/project1/deep_doppelganger/out_video","output_type":"window","resolution":"720p","parent_path":"/project1/deep_doppelganger"}' --output json --no-color ``` Expected: a Window COMP/output wrapper exists. - [ ] **Step 2: Arrange the network** Run: ```bash rtk node dist/cli/agent.js arrange --params '{"path":"/project1/deep_doppelganger","recursive":true}' --output json --no-color ``` Expected: nodes are laid out in readable zones. - [ ] **Step 3: Validate errors** Run: ```bash rtk node dist/cli/agent.js nodes errors --params '{"path":"/project1/deep_doppelganger","recursive":true}' --output json --no-color ``` Expected: no fatal errors on the final output path. Camera or MediaPipe permission warnings are reported honestly if present. - [ ] **Step 4: Capture preview** Run: ```bash rtk node dist/cli/agent.js preview /project1/deep_doppelganger/out_video --out _workspace/deep_doppelganger/out_video.png --no-color ``` Expected: `_workspace/deep_doppelganger/out_video.png` is written and is not blank. Note: on a licensed TouchDesigner system, the same graph can be promoted to `1920x1080`; the validated local target is `1280x720` because TouchDesigner Non-Commercial caps TOP output resolution. ## Self-Review - Spec coverage: bridge, tracking, visual entities, final output, fallback behavior, and layout validation are covered. - Red-flag scan: no vague implementation steps are used; every command is concrete. - Type consistency: command names and parameter names match schemas inspected from the local tdmcp-agent command catalog.