AI Agent Workflow Collection

Fan out a question to multiple independent AI agents (Gemini/Antigravity, OpenAI/Codex, and optionally a third model of your choice) through PAL's clink tool, then synthesize their answers into one recommendation. Use for multi-agent brainstorming, getting a second/third opinion on a design or plan, sanity-checking a decision across model families, or when the user says "ask the other AIs", "brainstorm with multiple models", or "get other perspectives".

The single home for delegating a bug hunt to clink agents — read it before sending any failure to clink-subagents or clink-brainstorm, not only a bug that already survived a pass. It routes a small bug to one cheap worker and escalates only what survives, defines the evidence that must travel with it, and enforces that no agent which produced a hypothesis may falsify or repair it. Use when a bug is about to be delegated, is still open after a worker looked at it, has two agents disagreeing about its cause, or came back after a fix landed.

What a master agent owns and can never delegate, and how to pick the model for a delegation from measured scores instead of memory. Use before any clink delegation, and whenever choosing between models, efforts, or clients — including "which model should review this", "is the cheap one good enough", "should I escalate the effort". Carries the full per-axis score table for every reachable model+effort inline, generated from the structured source, so the choice is made against data rather than recollection. Pairs with clink-subagents (handing out work), clink-brainstorm (convening judgment), and clink-debug (hunting a bug).

Delegate a well-scoped chunk of WORK (implementation, refactor, bulk transform, focused research, first-draft) to Codex (GPT-5.6) or Antigravity (Gemini) as a subagent via PAL's clink tool — to offload effort, parallelize, or save your context. Use when you have a self-contained, verifiable subtask a cheaper agent can do while you orchestrate. NOT for gathering opinions/consensus (use clink-brainstorm) and NOT for orchestration/judgment (keep that yourself). Covers which agent fits which task (grounded in Artificial Analysis index data + a local benchmark), how to prompt them, and the mandatory verify-everything discipline.

Web UX & Conversion Psychology framework based on Self-Made Web Designer (Chris McCoy) and Ran Segall (Flux Academy). Implements 3-Brain Persona Alignment (Survival, Emotional, Rational), Mental Model Layout Familiarity, MAYA Principle pattern breaks, Cognitive Chunking (3-4 item Working Memory Rule), Luxury White Space ("Space is Wealth"), Signature Niche Positioning, and the Oliver Smithies "Wander Off" experimental directive.

Universal micro-UI/UX design & engineering rules based on Kole Jain, Chris McCoy, Ran Segall, and Satori Graphics' frameworks. Enforces Pro Typography hacks (-2% tracking, Major Third 1.25x scale, 150% body line-height), 60-30-10 color balance, Material opacity modulation, 12/8/4 column grids, 8pt spacing rhythm, 4-state buttons, scrolling CTA frequency, progressive backdrop blur overlays, Luxury White Space, and the LIFT System with 6 Levels of Visual Flow.

One-time autonomous design context discovery. Scans the codebase to extract design tokens, typography, colors, spacing, brand patterns, and tech stack, then writes a Design Context section to CLAUDE.md. Zero questions asked. Use when: 'setup design', 'design context', 'design tokens', 'brand discovery', 'design system scan', 'initialize design', 'teach design'.

Use when the developer hands the agent a bounded batch to run unattended in a T4-team repo (T4 Labs / Slow-Inc) and steps away — "AFK", "run this while I'm away", "just handle it / take it from here", "clear the ready-for-agent queue", "keep going without me", or any long autonomous run where no human will approve each decision. Covers the preflight scope-lock, the may-decide-alone vs must-park boundary, the safe per-item loop, the stop-and-park conditions, and how to land the batch so the repo is never left broken or silently changed.

Use when working as the primary coding agent in a T4-team repo (T4 Labs / Slow-Inc) and you need durable memory across sessions and context-compaction — at session start (what's the open work, what did past sessions decide), after shipping or learning something worth persisting, or when open work is scattered across issues, ADRs, and MD files. Covers the team memory vault, the open-work ledger, the ship log, the survey-provenance cache, and Serena code memories. Triggers include "where did we leave off", "record what I shipped", "read the project memory", "what's still open".

Use when planning or implementing a feature/change in a T4-team repo (T4 Labs / Slow-Inc) — deciding how to go from idea to shipped code, filing or updating a GitHub issue or PRD, writing a bilingual (Thai + English) issue/PR body, opening a PR, or closing an issue. Covers the grill→PRD→issues→TDD pipeline, the PRD→issues→PR gate, the auto-triggered skill map, triage labels, and the issue lifecycle. Triggers include "let's build X", "file an issue for this", "write the PRD", "open a PR", "what labels", "how do we work here".

Use when working in a T4-team repo (T4 Labs / Slow-Inc) and something notable just happened that a future agent will need the "why" of — you fixed and validated a bug, made a hard-to-reverse architectural decision, shipped a system-affecting change, or hit a bug whose lesson is worth keeping. Helps pick the right record (post-mortem vs ADR vs system-impact entry vs bug-case-catalog) and write it so it stays a reliable index (file:line, commit SHAs, validated-only). Triggers include "write a post-mortem", "record this decision", "log this change", "should this be an ADR".

Use when starting, scaffolding, or setting up a new repository for the T4 team (T4 Labs / Slow-Inc) — or retrofitting an existing one — so an agent-primary repo gets the team's standard operating layer (domain glossary, product brief, decision records, durable memory, dev workflow) in one pass instead of hand-copying docs from a sibling repo. Triggers include "set up a new T4 project", "scaffold the docs/conventions", "add CLAUDE.md + docs/agents", "bootstrap the governance layer", "port the MangaDock conventions".

Use at the start of any task in a T4-team repo (T4 Labs / Slow-Inc) to decide which T4 skill applies — setting up the repo, session-start memory and open-work, planning or shipping a feature, or recording a bug/decision/change — and to hand off to the companion skill ecosystems the team runs alongside (superpowers, matt pocock, 9arm). The entry-point map for the t4- skill family; consult it before acting when you're unsure which T4 discipline or which external skill a task needs. Triggers include working in a repo whose CLAUDE.md points here, "how do we work in this repo", "which T4 skill", and starting a session in a Slow-Inc repo (MangaDock, T4-Fastwork).