searching-comfyui-solutions

Searches bundled templates and allowlisted web sources for ComfyUI workflow recipes with conservative ranking.

Updated Apr 8, 2026
One-click install
npx skills add https://github.com/ShinyGua/ComfyUI-Agent --skill searching-comfyui-solutions-shinygua
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: searching-comfyui-solutions
Source: https://github.com/ShinyGua/ComfyUI-Agent/tree/main/.claude/skills/searching-comfyui-solutions
Command: npx skills add https://github.com/ShinyGua/ComfyUI-Agent --skill searching-comfyui-solutions-shinygua

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? When a planner receives a ComfyUI request that does not map to any bundled capability-Skill goal, it has no grounded recipe to act on. This Skill retrieves matching workflow recipes from local templates and trusted web domains, then ranks them conservatively so the planner never anchors on a low-confidence result. ## Core Features & Use Cases - Local template search: Greps bundled Skill templates and reference folders for keyword matches, tagging hits with trust_tier "bundled" and the capability goal needed for deterministic planner routing. - Allowlisted web search: Fetches DuckDuckGo HTML results via stdlib urllib, filters them against policies/search_allowlist.yaml, and stamps each hit with trust_tier "web_trusted" and an ISO 8601 retrieved_at timestamp. - Conservative ranking: Merges local and web results, prefers bundled sources, and returns an empty list when the top score falls below the 0.3 threshold so callers fall back to asking the user. - Use Case: A user asks for an unfamiliar ComfyUI effect; the planner invokes this Skill, finds no confident recipe, and safely asks the user for clarification instead of guessing. ## Quick Start Ask the agent to search for a ComfyUI workflow recipe matching your request, for example by saying "search for a ComfyUI recipe for inpainting a masked region".

Frequently Asked Questions about searching-comfyui-solutions

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I search for ComfyUI workflow recipes from an agent planner?▼

Pipe a JSON payload with a query field into scripts/search_local.py for bundled template matches and scripts/search_web.py for allowlisted web results, then merge them with scripts/rank_results.py. Each script reads stdin and writes JSON to stdout.

How does the web search enforce trusted domains?▼

search_web.py loads allowed domains from policies/search_allowlist.yaml and only keeps DuckDuckGo results whose hostname matches the allowlist. Surviving hits are tagged trust_tier web_trusted with an ISO 8601 retrieved_at timestamp.

Why does the search return an empty results list?▼

The ranker applies a conservative threshold: if the top merged result scores below 0.3, it returns an empty list with reason below_threshold. This prevents the planner from anchoring on low-confidence recipes; callers should ask the user instead.

Does the web search require any third-party Python packages?▼

No, all scripts use only the Python standard library, including urllib.request for HTTP and html.parser for extracting DuckDuckGo result links. Network failures return a JSON error object rather than raising exceptions.

What are the trust tiers for search results?▼

Results carry one of three tiers: bundled for repo-shipped templates, web_trusted for allowlisted domains, and web_unverified for anything else. Only bundled and web_trusted results can skip the safety dialog during automatic execution.