Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add ChevalGrand520/local-gpu-imagegen --skill local-gpu-imagegengit clone --depth 1 https://github.com/ChevalGrand520/local-gpu-imagegenWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/chevalgrand520/local-gpu-imagegen/local-gpu-imagegen)<a href="https://agentmods.dev/skills/chevalgrand520/local-gpu-imagegen/local-gpu-imagegen"><img src="https://agentmods.dev/badge/skills/chevalgrand520/local-gpu-imagegen/local-gpu-imagegen/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/chevalgrand520/local-gpu-imagegen/local-gpu-imagegen"><img src="https://agentmods.dev/badge/skills/chevalgrand520/local-gpu-imagegen/local-gpu-imagegen.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00035 | $0.06530 |
| Opus 5 | $0.00017 | $0.03265 |
| Sonnet 5 | $0.00007 | $0.01306 |
| Haiku 4.5 | $0.00003 | $0.00653 |
Grade A, and why
local-gpu-imagegen scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 341 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Local GPU Visual Asset Workflow
Core Rule
Brief first, resolve one exact local route, confirm after displaying it, then run a bounded generate-review loop. Never substitute guesses for missing high-impact intent, model identity, or visual evidence.
The plugin exposes exactly seventeen MCP tools: local_gpu_imagegen_check, local_gpu_generate_image, local_gpu_discover_models, local_gpu_inspect_workflow, local_gpu_register_workflow, local_gpu_list_profiles, local_gpu_set_model_trust, local_gpu_recommend_models, local_gpu_start_run, local_gpu_get_run, local_gpu_branch_run, local_gpu_prepare_mask, local_gpu_confirm_mask, local_gpu_generate_round, local_gpu_record_review, local_gpu_finalize_run, and local_gpu_cleanup_run. Use the fifteen high-level discovery/onboarding/profile/run/revision tools for adaptive runs. The check and direct-generation tools are compatibility tools, not shortcuts around briefing, route resolution, and confirmation.
Runtime Recovery
Treat managed ComfyUI startup and model inventory as separate boundaries. A
healthy endpoint must pass both /system_stats and /queue; a port listener by
itself is not reusable. API discovery reads /object_info choices and does not
require a checkpoint to be loaded into VRAM. Never advise the user to load a
checkpoint merely because discovery failed.
- For
endpoint_unreachable, inspect the bounded backend logs and the returned endpoint, then retry only after the configured root and loopback endpoint are still exact. - For
startup_conflict, retry the same managed request. Another MCP instance may be starting the identical root and endpoint; do not launch ComfyUI manually, choose another port, or report missing models. - For
startup_process_exited, use the returned stderr summary and log paths to identify the startup failure, correct that failure, then retry startup. Do not wait out the full configured timeout after the child has exited. - For
api_inventory_failed, retry API inventory and inspect ComfyUI logs if it fails again. The endpoint is reachable, but model availability is unknown; do not call it an empty installation. - For
no_models_installed, install a supported model through the explicit bootstrap/download and license flow, then retry API inventory. Report this error only after every required/object_infoloader inventory succeeded and the checkpoint and split-model choices were all empty.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 341 lines · 35 tokens per session scan A 5659def39b54
local-gpu-imagegen is a skill published in the GitHub repository ChevalGrand520/local-gpu-imagegen (1 stars, last pushed 9d ago), licensed MIT. It adds 35 tokens to every session and 6,530 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
workflows
Knowledge about ComfyUI workflow API format, node connections, and common patterns. Use when helping users build, modify, or understand workflows.
workflow
Create a ComfyUI workflow from a template.
gen
Generate an image with ComfyUI from a text prompt.
history
Show recent ComfyUI generation history.
models
List available ComfyUI models, optionally filtered by type.
progress
Show execution progress for a ComfyUI job.