AI Agent Workflow Collection

Build and validate auditable image-production queues from action, outfit, prompt, or other combinable source assets. Use when jobs need stable IDs, deduplication, explicit references, diversity-aware batches, exclusion rules, and deterministic target paths before generation begins.

Build local UTF-8 HTML galleries for reviewing batch-generated images with grouping, numbering, prompt lookup, default exclusions, and image-load checks. Use when reviewers need a complete or batch view while controlling repeated outfits and retaining hidden historical outputs.

Use when an AI beginner is about to send, submit, or claim completion of a real document, file, image, slide deck, page, or other deliverable. Do not use for a trivial chat answer with no deliverable. Inspect the actual result, report the most important problems and unknowns, and avoid false completion.

Use when an AI beginner has a vague idea, missing key facts, an unclear goal, or asks what to do first. Do not use when the request is already specific and executable. Turn the idea into one small task, ask at most one blocking question, then proceed.

Safely remove advertisements, watermarks, website notices, contact lines, and other non-body residue from TXT, Markdown, OCR text, ebooks, transcripts, or copied articles without altering the original. Use when Codex needs to detect mixed or obfuscated domains/emails, repeated promotional lines, or clearly bounded attached artifacts; preserve UTF-8, GBK/CP936, GB18030, or UTF-16 encoding, BOM, and line endings; create a verified clean copy; and produce an auditable report. Use only on material the user owns or is authorized to process. Do not use for rewriting, censoring, paraphrasing, or modifying body content.

Compose and validate reference-guided image prompts that separate fixed visual dimensions from variable dimensions and acceptance criteria. Use when prompts must preserve an outfit, product, pose, or other reference evidence without copying irrelevant identity, scene, or unsupported details.

Reconcile batch image-generation results against local files, receipts, hashes, prompts, and queue records. Use when worker completion claims, missing images, duplicate generation, network retries, moderation blocks, or stale queue statuses must be resolved without loading image/Base64 responses.

Analyze visual-group concentration and record reversible human selection decisions for generated image collections. Use when repeated outfits, weak variants, rejected images, or default-hidden history must be controlled without deleting source files or confusing presentation choices with production facts.

Diagnose Windows proxy mismatches when a browser or ChatGPT can access the internet but Codex Desktop, Codex CLI, VS Code, or a terminal keeps reconnecting or cannot connect. Follow the fixed order: identify the active proxy IP and port, compare HTTPPROXY and HTTPSPROXY, determine whether already-open apps inherited stale values, verify from a newly started CMD context, then detect multiple proxy apps or conflicting ports. Use for read-only self-checks before changing proxy settings.

Used when a new ChatGPT conversation or Codex thread takes over an existing project. It first completes material assessment, follow-up interview, previous-conversation verification, and minimal-execution limits to prevent surface-level understanding from drifting away from the actual direction.

Create explicit worker packets for batch image generation while isolating image/Base64 responses from the controller. Use when multiple jobs or workers must receive exact IDs, prompts, references, output paths, receipt requirements, retry limits, and a prohibition on autonomous continuation.

Archive external AI-platform conversations such as Miaoxiang, music studio agent sessions, browser-based chat tools, and other logged-in web conversations. Use when the user wants to save, restore, visualize, structure, or hand off a long platform conversation while preserving read-only safety and avoiding credential capture.

Build human-facing visual entrances and thread-facing structured indexes for collected external materials, including videos, images, notes, comments, transcripts, screenshots, creator batches, liked lists, and conversation archives. Use when the user wants to visually inspect collected material or when other threads need a structured manifest.

Prepare and validate handoffs from external material intake to execution threads, project controllers, asset-writing threads, or analysis threads. Use when low-dispute repetitive tasks need delegation, when downloaded materials must be summarized for another thread, or when a receiving thread needs clear permissions, paths, boundaries, and expected return format.

Manage external material intake, download/collection routing, transcription decisions, visual/user-facing galleries, thread-facing structured indexes, and safe handoff for Douyin, Xiaohongshu, Bilibili, Miaoxiang, webpages, creator batches, comments, screenshots, transcripts, and other outside assets. Use when the user provides external links, creator profiles, short-video batches, platform conversations, comments, music/material references, or asks to download, collect, transcribe, visualize, classify, route, analyze lightly, or prepare materials for other project threads.

Collect and organize Xiaohongshu-style note platform materials: profiles, notes, images,正文, visible comments, likes, collects, comment counts, search samples, and semi-automatic logged-in browser collection. Use when the user provides Xiaohongshu profiles, note links, topic keywords, or asks for a visual and structured note archive without manual screenshot copying.

Handle external short-video materials such as Douyin, TikTok China, Bilibili mirrors, creator batches, liked lists, collections, downloaded videos, transcripts, screenshots, keyframes, comments, and routing summaries. Use when the user provides short-video links, asks to download only, transcribe, analyze lightly, collect comments, fix gallery grouping, or prepare video materials for later project threads.

Finalize a batch image-production delivery by reconciling queue records, local images, receipts, terminal blocks, exclusions, prompts, machine indexes, and a human-readable report. Use when production is ending or when completion claims conflict with local evidence.

Use when an AI result was rejected, visibly wrong, broken, unreadable, or failed to open or run, and the next repair is not already obvious. Do not use for an ordinary preference tweak or when the user has already specified the exact fix. Use evidence to choose the smallest useful repair instead of blind regeneration.

Use when an AI beginner explicitly wants another person, AI assistant, or new chat to continue an existing task. Do not use for internal task creation, background job control, or ordinary final delivery. Produce a short copyable handoff with the goal, materials, current result, next action, and anything that must not change.

Use when an AI beginner wants to create, reorganize, or clean up an everyday work slide deck or PowerPoint from notes, documents, an outline, or an existing deck. Do not use for a single poster, professional keynote art direction, or data analysis that must happen before the story is known. Produce a real deck when slide tools are available, keep the story clear and the visual system consistent, and verify the rendered slides.

Use when an AI beginner has existing notes or text that are messy, stiff, repetitive, jargon-heavy, or hard to understand. Do not use as the default for every writing request or when specialist legal, academic, brand, or technical writing is the main need. Rewrite for the intended reader while preserving facts.

Use when an AI beginner wants a simple poster, slide deck, page, or document layout to look clearer, less awkward, and presentable enough for everyday use. Do not use for professional brand systems, complex product UI, illustration, or video direction. Apply a small visual safety net without changing meaning.

Use when an AI beginner wants to create or restyle a simple poster, social card, one-page layout, document cover, simple page section, or other everyday static visual. Do not use for slide decks, professional brand systems, complex product UI, specialist illustration, or video. Choose one clear direction, create the actual result when tools allow, and keep it clean, readable, and free of generic template styling.