skills

Design a hyper-tailored, EVOLVABLE work loop for yourself — pick the right loop primitive, build an ungameable verifier, persistent state, and hard stop conditions, then add a bilevel reflect pass so the loop improves itself run over run. Use when setting up recurring/autonomous work (a nightly job, a watch-react monitor, a drain-until-done queue, an experiment loop), when asked to "keep doing X until Y", or when an existing loop stalls, overruns budget, or repeats mistakes.

Iteratively refine any visual output — charts, plots, web UIs, diagrams, slides, 3D renders — by rendering it to an image, actually looking at the image, critiquing it against the goal, and editing until it converges. Use when output quality is judged by eye ("make it look right", "fix the layout", "the legend covers the data", "polish this chart/page/diagram") or whenever generating visual output whose first attempt is unlikely to be final.

Recursive principal-subordinate orchestration for multi-agent / multi-PR work. Load when delegating substantive code, audit, or research work to a subordinate agent — OR when you ARE a subordinate that may spawn its own — OR when coordinating 2+ concurrent subordinates. Load BEFORE writing the brief, not after the subordinate comes back broken.

Methodology for running a rigorous, tiered computational-science investigation — DFT/MD/ab-initio/simulation campaigns that test competing hypotheses about a material, molecule, structure, or mechanism and end in a testable prediction. Load when planning or executing a multi-stage compute investigation (materials/chemistry/physics): enumerating hypotheses, choosing methods + compute tiers, gating input-model validity before production, separating mechanism from root cause, designing the confirming experiment, and writing claims without overclaiming. Distilled from a proven real-world investigation plan; domain-agnostic.

Deploy the current project to Vercel, GitHub Pages, or Netlify — detect the project type, verify credentials, build, and run the platform's deploy commands. Use when the user asks to deploy a site or check deployment/auth status.

Check and configure development-environment credentials and tools (GitHub CLI, git identity, Vercel/Netlify/Fly, npm/pnpm/bun, AWS/GCP/Azure) and report a status table with fix commands. Use when the user wants to verify or set up their dev tooling and logins.

Add captions to a talking-head video. ONE catalog (CATALOG.md) of 31 visual identities behind two engines: column-flow (captions composited INTO the scene — matte occlusion + mix-blend; cream/ink/editorial/keynote/documentary/loud/neon/glitch/chrome/velocity) and themed constitutions (anchor/ordnance/terminal/neonsign/stardust/stomp/scoreboard/transit/vhs/arcade/dossier/laser/thunder/hologram/biolume/aurora/spectrum/papercut/popup/chalkboard/graffiti/brush/inkwater/ransom/lastpage — e.g. a glyph-decode climax, a neon sign WRITTEN stroke by stroke, or the quiet `anchor` rail default). Route by identity, never by mode. Trigger on "captions/subtitles", "embed/cinematic captions", "VFX captions", "炸/特效/酷炫字幕", a named identity, or top-tier motion-graphics asks. Embedding every word is wrong for most talking-head content — `anchor` is the verbatim default. Pipeline: transcription → hyperframes remove-background matting → HTML render → ffmpeg overlay. Requires hyperframes and a single-subject clip.

Procedural high-fidelity CAD buildout of rooms / equipment / props in Autodesk Fusion, driven from WSL via the Fusion MCP. Load when you are about to script geometry into a Fusion Design — modeling a room, an instrument, a vessel/tank/kiln, casework, glassware, a fixture, or any real-world object — and the bar is COMPREHENSIVE detail (recessed door/drawer panels, knobs/displays/status-lights, spouts/necks/caps, 5-star bases, fume-hood sashes) rather than a single blocky primitive. Codifies the runtime model (def run(_context), cm-internal units, file read-back over the never-surfaced return value, no ui.messageBox), the into-component local-coord build pattern + Z##_Room_Element_NN naming + idempotency, the shape-decomposition vocabulary (primitives + the small features that make an object READ), the plan-spec-then-staged-build workflow, paste-ready API helpers (offset-start extrude, cyl, ring/annulus, loft cone, rotated-polygon leg, placement transforms), the verification loop (bbox read-back + containment a

Package an existing talking-head / interview / podcast video by layering timed, designed GRAPHIC OVERLAY cards onto the playing video — titles, lower-thirds, data callouts, quotes, side panels, picture-in-picture — synced to the transcript. The source video plays in full; the agent designs and writes each card's HTML in conversation, then renders to MP4 via hyperframes. Use when the user asks for graphic overlays, on-screen graphics / lower-thirds / data callouts / kinetic titles on a video, "package / dress up my video", "add overlay cards / graphic cards", or AI-composed graphic packaging of an existing video. NOT for plain subtitles (→ embedded-captions) or building a video from scratch (→ the creation workflows); when unsure overlays-vs-captions, see /hyperframes-read-first.

All animation knowledge for HyperFrames — atomic motion rules, multi-phase scene blueprints, scene transitions, broader motion-design techniques, AND the seven runtime adapters (GSAP default, plus Lottie, Three.js, Anime.js, CSS keyframes, Web Animations API, TypeGPU). Use for any motion or animation task: pick 2-4 rules and compose, or load a blueprint, or look up runtime-specific API (e.g. GSAP eases / Lottie player / Three.js mixer). HyperFrames-native: single paused timeline, seek-safe, deterministic.

Methodology + living tracker for systematic investigation, debugging, and validation work where a result contradicts expectation and you must descend into confounds, remediate, and climb back to the goal. Use when a result is surprising/wrong, when debugging a multi-layer problem, when reproducing/calibrating against a reference, or any time the work keeps spawning new sub-investigations that risk losing the thread. Keeps a route doc with a branch tree, findings log, and scorecard so progress is never lost across sessions.

Full-spectrum audit of a scientific manuscript for journal-submission readiness. Covers content accuracy (numbers, claims, data), narrative flow (paragraph-to-paragraph content sequencing), academic register, structural integrity, figure audit, citation hygiene, layout/presentation, submission hygiene, and reproducibility. Use when the user asks to "audit the manuscript", "review the paper", "fix up / polish / improve" a manuscript at the level of journal-submission quality, or asks "is this submission-ready / does this read like a [target venue] paper". Project-agnostic: works for any scientific manuscript in LaTeX, Markdown, or Word.

Geometry-first VMD pipeline for publication-quality molecular images: canonicalize the structure with PCA, generate candidate camera views on an azimuth/elevation/zoom grid, score them with objective visibility metrics, and render the best. Use when camera placement for a molecular scene must be deterministic rather than guessed.

Build polished, human-looking PowerPoint decks with python-pptx from an already-defined outline plus provided figures/tables/text. Use when asked to generate or update a .pptx programmatically with a consistent visual system, template-driven layout, and no generic AI-deck smell.

Run a bounded, evidence-first Technical De-Risking Compass (TDC) before committing to full development — or at any expensive, hard-to-reverse decision point. Load when a project needs its core premise or architecture tested with the cheapest experiments that can actually distinguish success from failure, BEFORE building the real thing. Triggers — "de-risk this", "is this feasible", "should we build it this way", "prove the architecture", "technical due diligence", "spike / proof of concept / feasibility study", "what could kill this project", "which assumptions are we betting on", "before we commit to <provider/architecture/hardware>". Produces falsifiable hypotheses with GREEN/YELLOW/RED/UNKNOWN verdicts, one thin end-to-end thread, a capability matrix, ADRs, and a risk-ordered roadmap. Domain-agnostic (software, AI/agents, scientific platforms, hardware, materials).

Render publication-quality molecular visualizations with vmd-python and Tachyon ray-tracing — environment setup, representations, materials, lighting, camera control, and batch rendering from Python. Use when producing molecular structure or trajectory images programmatically without the VMD GUI.