The Lab.
The Lab.
The lab started as ten custom GPTs — instruments, each tuned to one job. The platform retires them on 11 December 2026, and the work had already moved on. It lives in skills now: files a coding agent reads on the way into a task — when to use it, the method, what it hands back, and the excuses it makes before it skips a step. The ones I wrote are here to take.
Distils a spoken brain dump into a sharp prompt: the intent under the words, the parts, the cuts, and a depth pass that asks for the goal beyond the goal and a proof from outside the conversation.
Attacks a distilled idea before planning it, makes two or three architectures compete against the requirements, and returns a build brief with a milestone ladder and the kickoff prompt.
Runs the whole loop in one flow: dump, distil, one pause for the follow-up, scaffold, and the brief written into the new repo. Needs god and scaffold beside it.
Reviews motion by its frames, not its code: the twelve principles of animation as a pass/fail table, a frame cited for every row, one revision pass, a re-render.
A third model family on call: Google’s models through their command line, headless and read-only, for a second review after Codex, a sourced fact check, or a read of a whole folder at once.
Runs the Codex CLI from inside Claude Code — model, effort, sandbox, resume — and treats its answer as a peer’s, not an authority’s. Adapted from the community Codex skill pattern.
Reverse-engineers what a product does from public evidence — docs, demos, talks, reviews — into a capability map, a data contract and a bolt-on spec. Behaviour only: never binaries, never code, never anything behind a login.
The aerospace plain-English standard cut down to a checklist for runbooks, handoffs and error messages: one instruction per sentence, word limits, one word for one meaning, the warning before the step.
The same checklist written for OpenAI Codex. It installs to ~/.agents/skills/ instead; the repo has its agents/openai.yaml.
Maps how someone sees the world, from their writing or an interview: five axes, named traditions with a confidence on each, and the tensions between what they say and how they reason.
One hard read of a belief, fear or pattern: the hidden root, the reframe, the archetype in play, and a closing line. No comfort, no hedging.
The widest read: one signal taken through philosophy, psychology, esoterics, memetics and timing as an eight-section report. Missing data is named as missing, never invented.
Recommends films, books, music and video by fit to a person rather than popularity, and pre-screens a title before you give it an evening.
The board across every project: liveness first, then what is blocked on me, then which parked projects have reached their wake date.
Plain Markdown, written for Claude Code. A skill goes in ~/.claude/skills/<name>/SKILL.md; the two agents go in ~/.claude/agents/. Each file says “the user” where mine said my name — put your own words in the triggers. The gemini skill also calls a small helper script, agy-run.sh; it comes with the repo.
Only what I wrote is here, with one exception: codex is adapted from the community Codex skill pattern, and says so. The skills I use from other people stay with their authors; the repo’s README lists them and where to get them.
“Your psyche, narrated through the myths it already knows.”
Surface unconscious patterns by mapping your life to archetypal narratives. Best used when you feel stuck and want a story-level reframe, not a clinical one.
“A dream read that skips the dictionary.”
Feed it a dream; it returns symbolic analysis grounded in your context, not generic dream-symbol lookups. Works best with vivid, specific recall.
“Full-spectrum analysis of any narrative the world runs on.”
Give it a conspiracy theory, a political narrative, or a cultural myth. It returns a structured intelligence briefing — origin, psychological contagion mechanics, operational implications if true, and future trajectories. Forensic, not sensational.
“A taste engine that learns what you keep coming back to.”
Tell it your mood, a constraint, or a reference point. It profiles your taste across films, music, books, and podcasts, then recommends with reasoning — optimising for resonance, not popularity.
“Total-context self-audit. No blind spots left standing.”
A structured interrogation of your current life — goals, habits, relationships, contradictions. Designed for serious inner work when you want the full picture, not comfort.
“Reverse-engineer any site into buildable specs.”
Give it a URL or screenshot. It returns the typography, colour system, layout grid, and interaction patterns as a structured brief you can hand to a developer or use yourself.
“Turn a vague ask into a prompt that actually works.”
Describe what you want from an AI model. It rewrites your intent as a structured, high-fidelity prompt with the right constraints, format, and specificity.
“A filmmaking collaborator that speaks in shots, cuts, and light.”
Describe a scene or a visual intention. It returns precise technical direction — framing, lens choice, lighting setups, edit rhythm, sound design — drawn from real cinematography craft, not generic advice.
“Describe it. See it. One prompt to a finished image.”
Give it a brief — a scene, a concept, a vibe. It infers style, lighting, composition, and mood, builds an optimised image prompt, and generates the result. No back-and-forth; it fills in the gaps intelligently.
“Movement diagnosis without the waiting room.”
Describe a pain, restriction, or movement issue. It returns likely causes, differential considerations, and a progressive rehab protocol. Not a replacement for imaging — a first-pass triage.