Agent Claude Code
Shot-design specialist. Translates creative intent into cinematography shot layouts, motion grammars, visual direction templates, and organic imperfection specifications.
Agent Claude Code
Critique panelist focused on layout hierarchy, whitespace balance, graphic devices, and visual slop checks.
Agent Claude Code
Critique panelist focused on typography legibility, copy alignment, copy slop check, and WCAG contrast audit rules.
Agent Claude Code
Executes the generation stage from an approved plan. Chooses models within allowed policies and manages scratchpad compilation and generation runs.
Agent Claude Code
DEFAULT entry point for all brand material generation. Campaigns control plane execution, handling stop reasons by routing tasks.
Agent Claude Code
Critique panelist focused on product-truth fidelity for the active brand (any brand, not hard-coded).
Agent Claude Code
Translates high-level visual directions, creative intent, and brand aesthetics into high-fidelity, descriptive image generation prompts.
Agent Claude Code
Cultivates design philosophy, builds robust campaign briefs, drafts creative plan proposals targeting the rubric axes, and manages identity palette/typography/devices.
Agent Claude Code
Campaign dossier compiler. Synthesizes inputs from the critique panel, applies min-biased aggregation, compiles before/after instruction sets, and formats the campaign review packet.
Agent
Emergency fallback for material batches only when typed brand tools are unavailable. Uses python3 -m brandgen commands from repo root, records concrete artifact paths, and reports generated version ids.
Skill Claude CodeCodex
This skill encodes Emil Kowalski's philosophy on UI polish, component design, animation decisions, and the invisible details that make software feel great.
Instructions file CodexOpenCode
AGENTS.md instructions for velinussage/brand-gen, covering brand-gen agent contract, the hybrid pattern, never-edit rules, mutation verb table and files agents may freely edit.
Skill Claude CodeCodex
Ideate branded content — figure out what to say, who to say it to, and which visual format fits. Produces structured content briefs that feed directly into brand-gen pipeline as prompt seeds. Use when the user wants to create branded social content, carousel slides, editorial cards, educational materials, or needs…
Skill Claude CodeCodex
Detect whether brand-gen has a first-class abstraction for aesthetic experimentation. Traces aestheticdirectionbrief.variants from creation in aestheticcuration.py through planbuilder, promptassembly, scratchpad, and pipelinerunner. Reports whether branchid / parentbranchid (already declared in pipelinetypes.py) are…
Skill Claude CodeCodex
For each typed MaterialPlan field, trace whether it survives as STRUCTURE (not flattened to free-text prose) into GenerationScratchpad, into the rubric scorer (BrandScorer.forward), and back into iteration memory. Identify the lossy boundaries where structured intent becomes a string nobody can re-parse. Output a…
Skill Claude CodeCodex
Logo, wordmark, and lockup workflow for brand-gen. Use instead of the main brand-gen skill when the artifact is primarily a mark — logos, wordmarks, lockups, icons, or symbol explorations. Triggers on: "design a logo", "explore marks", "wordmark", "lockup", "icon design", "logo exploration", "brand mark", "symbol…
Skill Claude CodeCodex
Diagnose the actual feedback shape of the brand-gen pipeline. Identify whether any phase can request re-execution of an earlier phase, whether PipelineResult carries a candidate-set rather than a single result, and whether scoring output ever feeds back into route or plan within a single run. Output a state- machine…
Skill Claude CodeCodex
Full 6-phase brand material generation pipeline: prepare → plan → validate → generate → critique → evolve. Converts a multi-agent workflow into sequential instructions any single agent can follow. USE WHEN: generating brand materials, creating branded assets, running the brand pipeline, iterating on brand visuals…
Skill Claude CodeCodex
Audit the contract between pipeline phases (route → plan → validate → execute → review → evolve). Enumerate every Namespace field actually consumed via getattr(args, ...) per phase, compare against the typed PipelineRequest / MaterialPlan / GenerationScratchpad field sets, and identify "phantom" fields — ones…
Skill Claude CodeCodex
On-demand reference pack for brand-gen. Load this when you need model selection guidance, social/feed platform dimensions, workspace file layout, or aspect ratio gotchas. Use when the agent needs to choose between models, pick the right image dimensions for a platform, understand where files live on disk, or debug…
Skill Claude CodeCodex
Setup and installation skill for brand-gen. Use when the user wants to clone, install, configure, validate, or connect brand-gen to an agent host for the first time. Covers Python environment setup, .env configuration, agent-browser installation, MCP registration, and host-specific notes for Claude Code, Pi, OpenClaw…
Skill Claude CodeCodex
Audit every prompt-truncation, compression, and lossy-slice site in the brand-gen Python package. Produces a manifest of every cap call (captextatsentence, compresspromptbody, evicttobudget, hard [:N] slice) with file:line, the limit value, the limit source (constant, JSON, hard-coded literal), and whether the drop is…
Skill Claude CodeCodex
Catalog every emitter of a quality verdict in brand-gen — visualreviewstatus, axisscores, vlmapproved, critiquepolicy decisions, render verification, generationflow auto-critic. Map each to a single proposed Verdict schema. Flag every place where verdicts are silently combined into a single string field (e.g.…
Skill Claude CodeCodex
Main skill for brand-gen — the file-backed brand-material runtime. Use for almost every brand-material session: onboarding into the correct workspace, choosing saved-brand vs testing session flows, planning assets or sets, running the pipeline, reviewing outputs, scoring, iterating, messaging ideation, inspiration…