Install
$ agentstack add skill-hybridlabor-api-bdb-dev-optimized-agent-skills-touchdesigner-self-debug ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ● Filesystem access Used
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ● Dynamic code execution Used
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
TouchDesigner Startup + Functional Self-Debugging
A general technique for verifying any TD-side Python change (e.g. anything in td/modules/**) inside real TouchDesigner. Calling functions directly from the Textport, without standing up the HTTP WebServer, is the shortest path.
> This skill is feature-agnostic. The worked example in Steps 4–5 happens to > verify node auto-alignment (the PR that produced this skill), but that is > illustration only. The reusable shape is: create/modify via the real code > path → read back the TD runtime value you actually care about → reconcile > against an invariant. Swap the alignment-specific bits (grid, overlap area, > node_layout.py) for whatever your change touches. Do not assume alignment, > grids, or overlap have anything to do with your task.
When to Execute
- You modified TD-side logic (e.g.
td/modules/mcp/services/*.py) and want to check behavior on the real machine. - Unit/offline verification is insufficient and you need to reconcile against TD runtime data (any node attribute, cook result, error state, etc.).
- You are in a fresh worktree that cannot rely on CI or the 9981-dependent integration suite.
Prerequisites and Pitfalls (read first — these are general)
- The
.toxreadstd/modules/from disk at runtime. Re-authoring the.toxis not required for.pychanges; reload modules or restart TD. loadTox()leavesexternaltoxempty →import_modules.setup()fails withNo module named 'mcp'. Replicate setup manually (Step 3).- A fresh worktree lacks generated artifacts → empty OpenAPI schema / missing
node_modulesbreaks HTTP routing. Bypass it by calling the Textport directly (Step 4). Only runnpm install && npm run genif you specifically need the HTTP path (e.g. running the vitest integration suite). - Computer-use screenshots go black when TD loses frontmost focus → recover with
open_application("TouchDesigner")+switch_display. - Multi-line pastes into the REPL break on line boundaries → always wrap in
exec("""...""")to submit as one statement. - To pick up a later
.pyedit in an already-running TD, drop the module fromsys.modulesso the next import reloads from disk (Step 3c). If the WebServer is already up you can do this viaexecute_python_script— no computer-use needed.
Step 1: Start TD (skip if already running)
computer-use: request_access(["TouchDesigner"])
computer-use: open_application("TouchDesigner") # wait a few seconds
computer-use: screenshot # confirm
If a WebServer is already listening on 9981 (lsof -nP -iTCP:9981 -sTCP:LISTEN) and get_td_info succeeds, you can drive everything via execute_python_script and skip the computer-use / Textport steps entirely.
Step 2: Open the Textport
Menu bar Dialogs → "Textport and DATs". (Alt+T toggles the Palette and is unreliable; use the menu.)
Step 3: Load .tox + manually replicate module setup
# 3a. Load
p = op('/project1').loadTox('/td/mcp_webserver_base.tox'); print('LOADED', p)
# 3b. externaltox is empty after loadTox — replicate import_modules.setup()
exec("""
import sys, os, yaml
b = op('/project1/mcp_webserver_base')
b.par.externaltox = '/td/mcp_webserver_base.tox'
mp = '/td/modules'; tsp = os.path.join(mp, 'td_server')
[sys.modules.pop(m) for m in list(sys.modules) if m.split('.',1)[0] in ('mcp','utils')]
[sys.path.remove(x) for x in (tsp, mp) if x in sys.path]
sys.path.insert(0, mp); sys.path.insert(0, tsp)
import mcp
sp = os.path.join(mp, 'td_server', 'openapi_server', 'openapi', 'openapi.yaml')
mcp.openapi_schema = yaml.safe_load(open(sp)) if os.path.exists(sp) else {}
print('SETUP_OK', mcp.__file__, 'schema', bool(mcp.openapi_schema))
""")
# 3c. Reload just your edited modules to pick up later .py edits (targeted, safe)
import sys
for m in ('mcp.services.api_service',): # list the modules you changed
sys.modules.pop(m, None)
SETUP_OK .../td/modules/mcp/__init__.py means mcp imported from the local tree. (schema False is fine for the Textport-direct path.)
Step 4: Verify your change from the Textport (no HTTP required)
Generic template — this is the part to reuse. Drive the code you changed through its real entry point, then read back the runtime value and reconcile:
exec("""
from mcp.services. import # the code you changed
# 1. exercise it via the real call path
= ()
# 2. read back the ACTUAL TD runtime value(s) you care about
=
# 3. reconcile against an invariant (prefer a numeric/boolean check over "looks right")
print('JUDGE', 'PASS' if else 'FAIL', )
""")
Notes that generalize:
- Pass
node_typeas a string (e.g.'circleTOP') toapi_service.create_node; TD resolves it. - Prefer reconciliation (compute a value and compare) over assertions that can be gamed.
- To test cumulative behavior, apply the change repeatedly and check an invariant like
after == before + N.
Worked example (align-nodes PR — illustration, not the procedure)
Verifying that create_node places nodes on a non-overlapping grid. The alignment/overlap specifics here are the feature under test, not part of the generic workflow:
exec("""
import td
from mcp.services.api_service import api_service
_t = op('/project1/_selfdebug');
if _t: _t.destroy()
test = op('/project1').create(td.baseCOMP, '_selfdebug')
[api_service.create_node(test.path, 'circleTOP', 'c%02d'%i, None) for i in range(12)]
boxes = [(c.nodeX, c.nodeY, c.nodeWidth, c.nodeHeight) for c in test.children]
ov = lambda a,b: max(0, min(a[0]+a[2],b[0]+b[2])-max(a[0],b[0]))*max(0, min(a[1]+a[3],b[1]+b[3])-max(a[1],b[1]))
tot = sum(ov(boxes[i],boxes[j]) for i in range(len(boxes)) for j in range(i+1,len(boxes)))
print('JUDGE', 'PASS' if tot==0 and len(boxes)==12 else 'FAIL', 'overlap=%s' % tot)
""")
Here the invariant is "total overlap area == 0". Your feature will have a different invariant entirely.
Step 5: Optional — a concurrent "offline judge" (no TD, deterministic)
If the change has a pure logic kernel that does not need the TD runtime, extract it into a module that does not import td, then verify it with system python3 by loading the file directly via importlib (zero extra deps):
# generic: load a td-free helper by path, bypassing the mcp package __init__
import importlib.util, os, sys
spec = importlib.util.spec_from_file_location(
"mod", os.path.join(sys.argv[1], "td/modules/mcp/services/.py"))
mod = importlib.util.module_from_spec(spec); spec.loader.exec_module(mod)
# ... call the pure functions, reconcile numerically, set exit code on PASS/FAIL
(In the align-nodes PR the pure module was node_layout.py; yours may not exist or may be named anything.) TD-side judging confirms integration; the offline judge proves invariants. Use both when the kernel is separable.
Cleanup
- Any scratch container you create (e.g.
/project1/_selfdebug) is discarded if you close an unsaved project. Delete explicitly withop('/project1/_selfdebug').destroy(). - You may leave TD open for reuse.
Related
- Global memory:
loadtox-externaltox-gotcha,fresh-worktree-needs-gen-for-schema,tox-embeds-import-modules-dat - HTTP path (only when needed):
npm install && npm run gen, then the vitest integration suite hits the live WebServer on 9981.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: hybridlabor-api
- Source: hybridlabor-api/bdb-dev-optimized-agent-skills
- License: Apache-2.0
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.