Design & Motion · LiamGvchi/gc-minimal-zine-poster
gc-minimal-zine-poster
Generate or analyze poetic paper-texture minimal zine posters with large negative space, a small editorial collage or focal element, experimental typography, and one clear color accent. Use when the user gives a theme, sentence, article, mood, object, photo, content brief, reference images, or an image folder and wants a final poster image, a production-ready image prompt, reusable style rules, or varied layouts that stay in one visual family.
L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.
1How It Works & Mechanism
The repository is purely orchestrated via prompts and reference documents with zero executable code: all capabilities stem from routing and execution instructions in SKILL.md plus layered knowledge across five 'runtime manuals' in references/; pixel rendering is delegated to the host's built-in image generation tool.
README.md
The Skill package contains no scripts, external fonts, API keys, private paths, or downloaded runtime assets. Image-generation quality and photo preservation still depend on the model available in the host environment.
The README Requirements section demonstrates zero scripts and zero assets; git ls-files shows the full file list (SKILL.md, README×3, LICENSE, agents/openai.yaml, references×5, evals/evals.json, examples×6 jpeg) contains no executable files like .py/.sh/.js/.ts—understanding that 'the skill itself contains no native generation capabilities' is the primary prerequisite for this repository.
Five-mode routing defaulting to the minimal viable mode: requests are mapped to Generate (default) / Photo Input (a Generate sub-flow) / Reference Analysis / Prompt-only / Analyze + Generate, autonomously making safe decisions without questioning the user.
SKILL.md
**Generate Mode — default:** theme, sentence, article, object, mood, photo, or brief → visual metaphor → prompt → generated image → inspection.
The 'Route The Request' section maps each request pattern to its minimal mode; 'If the intent is ambiguous but clearly asks to “做一张”, use Generate Mode. Do not stop to ask about choices the skill can make safely.'—designed to minimize user interruptions.
On-demand knowledge loading: style-system is required for every mode, while remaining reference files are loaded only in their corresponding modes to avoid diluting the execution context with irrelevant instructions.
SKILL.md
Read `references/style-system.md` for every mode.
'Load The Relevant References' defines five criteria corresponding to when each reference file should be read (prompt-compiler only for Generate/Prompt-only; reference-analysis only when reference images are provided, etc.)—a layered architecture of a thin SKILL.md and rich references.
Photo inputs are formalized into a three-step protocol of 'Role Classification + Retention Level + Actual Image Forwarding': each image is assigned an edit target / reference image / supporting insert role with a High/Medium/Low retention tier, and the actual image (not a text description) is passed to the generation tool.
SKILL.md
Pass the actual input image to the built-in image-generation tool. Use `referenced_image_paths` when every target image has a local path. Use `num_last_images_to_include` when at least one target exists only in the conversation, choosing the smallest number that includes every target, up to five. Never use both mechanisms in one call.
This level of formalization is rare among 'photo-to-poster' skills: recognizable people/pets/products default to High retention with 'Prefer an original-photo crop, clipping, or printed fragment over redrawing'; two mutually exclusive image transfer mechanisms (max 5 images) embed the host tool's real capability boundaries directly into the workflow.
Prompt compilation is formalized to near-executable precision: a 10-field ordering + a 4-section body template + fill-in placeholder blocks, with explicit bans on unconstrained adjectives and 'unnamed abstract compositional words' (a regression fix in v0.3.1).
references/prompt-compiler.md
Do not recite the field names. Write decisive image directions as four compact paragraphs.
The template slots negative space percentages, cluster position and size ratios, exact high-chroma hue, and its physical medium (photo crop / ink block / specimen / printed illustration / texture window / typography); Compilation Rules strictly dictate 'State actual placement and size. Avoid words such as `somewhere`, `minimal`, `nice`, `artistic`, or `balanced`'—rejecting vague phrasing that models cannot translate into pixels.
A variation engine providing 'grammar-level variation within the same visual family': batch outputs must alter grammatical axes like layout families, focal forms, and typography distributions, rather than just nudging coordinates or repeating safe defaults.
references/variation-engine.md
Change visual grammar, not only coordinates. A new recipe should alter at least layout family, focal-element form, and typography mode relative to the preceding visible output.
A discrete option pool: 10 layout families × 10 focal-element forms × 8 typography modes × 8 texture modes × 0–3 ornament pieces; SKILL.md Variation Discipline bans defaulting to 'tiny centered photo + blue dots + microtext', logging recipes in a fixed sequence for auditability.
Mandatory post-generation quality gate: generated images must pass a self-audit of 12+ focal checklist items before delivery; if failed, a single prompt-tightened regeneration is authorized, and persistent failures must explicitly report limitations rather than passing off defective outputs as successes.
references/quality-gate.md
If one of the central checks fails, revise the prompt and regenerate once.
Focal checks verify: 3:5 aspect ratio, 70%–90% negative space, single visual event occupying 8%–25%, thumbnail-visible dominant accent (0.8%–2.5% canvas or 15%–35% cluster), 'No unrequested maritime symbol, ship-mooring icon, or generic pictogram has been introduced', avoiding copying reference text/layouts, etc.; SKILL.md Generate Step 6 reiterates these checks and adds retention level verification for Photo Input.
2Core Capabilities
01End-to-end generation: Content → Visual Metaphor → Image Generation Prompt → Host Image Rendering → Quality Gate Check → Standardized Delivery (Generated Image + Final Prompt + Recipe + Explanation).
02Photo input handling: Three-role classification + High/Medium/Low retention tiers + actual image forwarding (referenced_image_paths / num_last_images_to_include, mutually exclusive, up to 5 images).
03Evidence-based style analysis of reference images: Separation of observed/inferred properties, promoting features to fixed rules only upon repetition, treating 1–2 instance features as sample residue without overgeneralizing, outputting Fixed/Variable/Residue sections + reusable prompt + confidence score.
04Prompt-only mode with strict boundary adherence: Outputs solely a four-section prompt + recipe + negative constraints, explicitly prohibiting any implication that an image was generated.
05Batch/multi-image variation discipline: ≥4 images must use at least 3 layout families; adjacent outputs must not repeat the same layout + focal combination; ≥60% of images must carry the dominant accent on the subject/color block rather than micro-dots.
06Source protection and anti-plagiarism: Strictly prohibits copying reference text, branding, watermarks, signatures, dates/locations, and exact compositions, preserving the underlying system rather than sample residue.
07Quality gate self-audit: Dual-level check (full view + thumbnail) verifying negative space, single visual event, primary color visibility, absence of unrequested maritime symbols/generic pictograms, and prevention of drift into commercial ad layouts.
08Built-in Chinese regression evaluation suite: 8 evals (including a v0.3.1 regression test ensuring 'non-maritime theme + bright red focal element must not introduce icons').
4Recommended Governance & Evidence
Suggested Controls for Safe Operation
For Agent Runtime
Restrict working directory
For Administrators
Retain operation logs
Evidence & Triggers (File Paths & Excerpts)
- 本地文件: Operate local project files or working directory artifacts
Audit Notes: The entire package contains zero executable code (no scripts, no external network endpoints, no credential access, no file writes triggered by SKILL instructions), functioning purely as a prompt and 5-reference document orchestration; at runtime, it only 'reads' user-supplied images/reference files (locally or within session) and drives the host's built-in image generation and inspection tools. The only 'outbound transmission' occurs when the host tool forwards the image prompt (and original user images during Photo Input) to the configured image backend—a process triggered only upon explicit user request, representing standard host tool behavior (e.g., Codex) rather than an action inherent to the skill. Key considerations are limited to three: generation relies on host model quality and licensing, user photos are uploaded to the host image backend with necessary host-side visibility, and the prompt injection surface is strictly conversational, handled entirely by the host.
5Second-Pass Verification
- [discrepancy] SKILL.md frontmatter name 与任务/仓库名一致性 — frontmatter name 为 gc-minimal-zine-poster-v0-3（callable name，README 明示向后兼容保留 v0.3 名），任务输出名/仓库名为 gc-minimal-zine-poster；指向同一 skill 无歧义，Analysis 按任务名 gc-minimal-zine-poster 入库（skill.name = 任务名，skill.path = 仓库根）。
- [ok] 全仓仅一份 SKILL.md 且位于仓库根 — git ls-files 17 项 + glob 复核：仅根 SKILL.md，无第二份 skill 清单或嵌套 SKILL.md。skill.path 记 '.'。
- [ok] 零脚本/零自有网络/零凭证/零写入 — 文件清单全为 md/yaml/json/jpeg/LICENSE；token 扫描零命中；.gitignore 中 outputs/、generated_images/ 仅预留宿主出图落盘目录，SKILL.md 无任何文件写入指令。
- [ok] 能力声明与源码相符（无夸大） — SKILL.md 逐字含 'Use the built-in image-generation capability.'；README 'Image-generation quality and photo preservation still depend on the model available in the host environment.' 明示依赖宿主——生成能力不是仓库自带。
- [ok] 五模式路由 / Load-the-References 机制 — 'Route The Request'、'Load The Relevant References' 两节原文复核命中，五条参考装载指令与 references/ 五个文件一一对应。
- [ok] 照片三角色 + 三保留级 + 真实传图机制 — SKILL.md Photo Input Mode 原文复核：三角色定义、'Choose and record one preservation level' 高/中/低三段、传图二机制（referenced_image_paths / num_last_images_to_include，上限 five）均逐字存在。
- [ok] prompt-compiler 字段顺序/四段式/禁模糊词 — references/prompt-compiler.md 复核：10 项 Field Order、'Four-paragraph Prompt Shape'、'Avoid words such as `somewhere`, `minimal`, `nice`, `artistic`, or `balanced`' 均在。
- [ok] variation-engine 布局族/批量纪律 — 10 个 layout family 列表与 'For a batch of four or more, use at least three layout families.' 原文复核命中。
6Verdict & Scenarios
Best for: Ideal for creators and social media publishers needing batch minimalist paper-textured posters within the 'same visual family' (converting themes, short copy, articles, or moods into posters requiring negative space, a single visual event, and a single high-chroma accent), designers needing to formalize reference images into reusable style rules, and lightweight image editing workflows transforming real photos into posters while preserving the identity of people/products (relies on a capable host image model). Friendly to Chinese inputs and outputs (example requests in SKILL.md, output templates, and the 8 evals are predominantly in Chinese).
Not for: Not suitable for hosts lacking built-in image generation capabilities (Generate and Analyze + Generate modes cannot be fulfilled); not suitable for posters requiring pixel-level precision or dense, long-form typography (models distort long text; extended words can only serve as texture); not suitable for workflows requiring editable layered source files or commercial print-ready production—deliverables are AI raster images + prompts + recipes; not suitable for users seeking to treat reference images as an 'editing source' to directly copy compositions or text (strictly forbidden by the skill).
Install Agent-Ready
① Verified Mirror 2026-09-09
Option A · Direct DownloadDownload gc-minimal-zine-poster.tar.gz
sha256:
Option B · Copy Guide for Agent7eaa9676cd748aba…Agents automatically download and verify mirrors from this JSON guide.
② Upstream GitHub
GitHub Repo ↗
Pinned to commit ddb0d66b24.