ANS VibeHub Skill & Workflow Matrix
Turn rough ideas into precise Agent briefs & reusable Skills
Write structured project briefs, rewrite vague wording into accurate technical terms, and copy battle-tested prompts from ANS’s 10 core daily Skills across engineering, media production, and de-slop writing.
For posters, storyboards, short videos, covers, and campaign assets.
Four-part verifiable brief builder
Saved on this deviceFunction 01 · Precision Expression & Proactive Term Hints
Turn everyday wording into unambiguous Agent instructions
Preserve your intent while adding only the 1–3 technical terms that remove ambiguity—never inventing frameworks, parameters, or unrequested features.
“Change the button color when I move the mouse over it, and make it react smoothly when I press it.”
Add Hover, Active, and keyboard Focus states to the primary button, and apply a 160ms Transition so the color and subtle scale shift feel smooth without changing button dimensions.
Adds only the exact state terms (Hover / Active / Focus / Transition) without inventing an animation library or redesigning the component.Function 02 · ANS 10 Core Daily Skills Across 3 Production Pipelines
10 high-frequency Agent Skills for engineering, video pipelines, and clean writing
Instead of installing dozens of overlapping global tools, chain focused project Skills into three repeatable pipelines: Engineering & Git Delivery, Talking-Head Video Production, and De-Slop Visual Documents.
Engineering, Sub-Agent Review & Git Delivery Pipeline
Map file paths with a fast explorer, execute bounded edits, run a context-free pre-commit simplification pass, and ship verified commits with a visual diff walkthrough.
- 01 Explore paths
- 02 Bounded implementation
- 03 Simplify code review
- 04 Ship & visual diff
Talking-Head Video & Build-in-Public Media Pipeline
Start from a 10-minute speaking outline and single-take screen recording, then chain VAD+ASR retake removal, screen-OCR subtitle correction, 16:9 + 3:4 covers, and draft-only uploads.
- 01 Pain-point outline
- 02 VAD+ASR rough cut
- 03 OCR captions & dual covers
- 04 Draft-only validation
De-Slop Writing, Visual Explainer & Slide Deck Pipeline
Strip 'not X, but Y' framing and inflated themes first, then turn dense technical notes into interactive single-page visual explainers, per-page HTML slide decks, or clean repository READMEs.
- 01 Fact extraction & de-slop
- 02 Visual architecture explainer
- 03 Per-page HTML deck
- 04 Bilingual repo showcase
01ans-team-modeEngineering, Sub-Agent Review & Git Delivery Pipeline
Multi-Role Sub-Agent Team & Cost Routing
Run this task in ans-team-mode: 1) dispatch an Explorer sub-agent to map relevant file paths and interfaces and return a compact summary; 2) specify exact rule files and target files before handing implementation to an Executor; 3) invoke the top-tier Advisor only if an architectural blocker persists across two rounds.
02ans-code-simplifyEngineering, Sub-Agent Review & Git Delivery Pipeline
Context-Free Pre-Commit Code & UI Simplification
Spawn a fresh context-free review sub-agent (ans-code-simplify) to audit our uncommitted changes: remove redundant defensive branches, duplicate derived state, leftover migration code, and exposed internal IDs or filler UI helper text.
03ans-git-shipEngineering, Sub-Agent Review & Git Delivery Pipeline
Branch Isolation, Clean Commits & Visual Diff Ship
Run the ans-git-ship checklist on the current working tree: 1) verify .gitignore excludes temp files and secrets; 2) run typecheck and unit tests; 3) create clean intent-scoped commits and summarize module boundaries and key diffs.
Function 03 · Building Agent Skills Like a Product
Six engineering rules that make Skills reliable even on lightweight models
Good Skills are designed for how models actually behave under long context and tight token budgets—separating deterministic scripts from flexible judgment and verifying quality with blind sub-agent reviews.
Run with Defaults First + Silent Post-Check Config Hints
Common pitfall：Forcing the Agent to inspect config files or read hundreds of lines of script source before every run wastes tokens and latency.
ANS productized rule：Let the Agent run the script immediately with sensible defaults; have the script print a brief footer note about optional customization, prompting setup only when a required credential is missing.
Deterministic Steps in Code, Flexible Judgment in the Agent
Common pitfall：Asking an LLM to calculate timestamps, concatenate files, or measure DOM geometry by hand is slow and error-prone.
ANS productized rule：Put every rule-bound transformation into a Python or Node script, and reserve semantic judgment, visual direction, and copy editing for the Agent.
Per-Page Intermediate Files So Lightweight Models Never Truncate
Common pitfall：When lightweight models generate a thousand-line HTML file in one shot, they frequently truncate later sections or silently drop source details.
ANS productized rule：Write each slide or stage to its own small intermediate file first, then use a deterministic build script to validate and bundle the final artifact.
Inject into a Local HTML Preview for Instant Human Review
Common pitfall：Reading raw JSON, SRT text, or layout descriptions in a terminal makes it hard for humans to judge visual quality or alignment.
ANS productized rule：Render candidate UI directions, video covers, or subtitle edits into a local HTML preview page with one-click copyable selection prompts.
Context-Free Sub-Agent Blind A/B Evaluation
Common pitfall：Asking the same Agent that wrote the code or Skill to grade its own output produces biased self-praise.
ANS productized rule：Run the same task with and without the Skill, then pass both outputs (or screenshots) to a fresh sub-agent for blind rubric scoring.
Project-Scoped Skill Isolation + AGENTS.md Entry Routing
Common pitfall：Installing dozens of domain-specific Skills globally bloats startup context and causes models to trigger the wrong tool when descriptions overlap.
ANS productized rule：Keep only universal Skills global; store domain workflows inside the project directory and route them cleanly from a concise AGENTS.md entry file.