GiggleLiu

Use when the user asks for a weekly advisor check, student TODO follow-up, research supervision update, or current external updates for a zlp-harness Zulip stream.

Use when creating or scaffolding a Cryochamber weekly advisor agent for a research harness, especially requests for weekly advisor cryo, cryo advisor, scheduled advisor agent, or adapting benchmark.harness.

Use when adding one or many new references (arXiv ID or DOI) to this project's `.knowledge/` library — fetches metadata via Semantic Scholar, downloads the arXiv preprint PDF, renders to markdown, and regenerates `INDEX.md`. Manifest is array-based; batches of 50+ refs work fine.

Use when the user wants to scaffold a new research-discussion harness: a private GitHub repo wired to a Zulip stream, with .knowledge for papers and a Makefile for the bridge. Triggers on create a new harness, scaffold a harness, init harness for a topic, /init-harness, or set up a harness like qec.harness for another topic. Self-contained: bundles the canonical Makefile, CLAUDE.md template, and a thin project-level onboard skill that wires up the zlp-harness plugin; does not depend on any source repo.

Use when a new collaborator on the <<TOPIC>>.harness repo just cloned and is running it for the first time. Enables the zlp-harness plugin in ~/.claude/settings.json on the first invocation, then on the second invocation (after an agent-client restart) delegates to the plugin's zlp-onboard skill for personal-account setup. Triggers on "I just cloned this", "first time setup", "onboard me", "help me get started", "/onboard".

Use when a new collaborator on any zlp-harness-based repo needs to bootstrap their machine — checks what's already installed, walks them through getting a Zulip API key for the harness's site, creates the global zlp-harness workspace directory, places `zuliprc` there, and verifies the bridge works. Triggers on "I just cloned this", "first time setup", "onboard me", "help me get started", "set up zulip", "/zlp-onboard", or when invoked by a harness's project-level onboard skill.

Use when the user wants to check the project's Zulip stream and usually draft a reply: pulls new messages, builds context from the project library, brainstorms reply framings, and sends only after explicit approval. Triggers on phrases like check zulip, any new messages, what did someone say, reply to that, /zulip, or /zulip-reply.

Use when a Cryochamber agent needs to send a message through the local mailbox outbox, including replies to mailbox inbox messages or proactive weekly Zulip advisor posts.

Use when the user wants to create a new cryochamber application, set up a scheduled agent task, or scaffold a cryo project with plan.md and cryo.toml

Use when the user wants a screenshot of the cryohub web dashboard for documentation, README, mdbook, or a slide deck. Spawns cryohub on a chamber workspace, drives headless Chrome, and saves a PNG to the requested path.

Use when preparing a new crate release, bumping versions, or tagging a release

Use when the user wants visual state machines of the cryochamber daemon — the overall event loop around inbox messages and TODO-triggered wakes, plus per-item lifecycles for a message and for a reminder — rendered as hand-written SVGs

Use when a PR has review comments to address, CI failures to fix, or codecov coverage gaps to resolve

Use after implementing a feature or any code change to verify completeness and correctness before committing

Use when creating or managing agent profiles for testing — guides users through selecting a feature/skill, defining a use case with expected outcome, and choosing or generating an agent persona, saving reusable profiles to docs/agent-profiles/

Use when testing a project's features from a user's perspective — simulates downstream users who read docs, set up the project, and exercise features end-to-end, reporting on discoverability, functionality, and documentation quality

Use when testing a paper's writing quality from a reader's perspective — simulates first-time readers with calibrated audience profiles who scan, then deep-read section by section, reporting on clarity, confusion points, structure, and flow

Use when testing a skill — role-plays through a skill's conversational flow with a simulated user persona, then produces a test report with structural analysis, UX feedback, and actionable suggestions

Use when the user wants to present or inspect STL models in the browser (a single self-contained HTML viewer page, works offline) and/or physically simulate a ball rolling through printed geometry (marble runs, tracks, funnels, tunnels) with the trajectory animated inside the viewer. Complements design-stl (static PNG previews + printability checks) with interactive 3D and dynamic validation.

Use when the user wants to slice an STL and print it on a Bambu Lab printer over the local network (LAN), preview an STL in 3D in the browser, or check/monitor a Bambu print's status. Covers slicing with Bambu Studio or OrcaSlicer, uploading over LAN, starting a print behind a confirmation gate, and streaming progress.

Use when setting up the Bambu LAN printing workflow for the first time on a machine, or when print-to-bambu preflight is not all green — installs/locates the slicer, sets up the Python venv, scaffolds bambu.toml, helps find the printer IP/serial/access code, and walks through enabling Developer/LAN Mode so prints can start.

Use when the user wants to create or design a 3D-printable STL model from a description (e.g. a molecule, a part, a parametric shape) rather than starting from an existing STL. Generates geometry with Python/trimesh (default) or OpenSCAD, renders previews, validates printability, and iterates. Pairs with print-to-bambu to actually print it.

Use when preparing a new yao-rs release, bumping the version, or tagging a release

Use when you have a GitHub issue and want to create a PR with an implementation plan that triggers automated execution

Use when a PR has review comments to address, CI failures to fix, or codecov coverage gaps to resolve

Use after implementing a model, rule, or any code change to verify completeness and correctness before committing

Use when Codex needs to animate Typst pixel-art character functions into a looping GIF, especially in pixel-family-style packages with 16x16 inline vector sprites, generated frame sheets, or ImageMagick/Typst GIF assembly.

Design a new pixel-family character interactively. Use when the user wants to add a new character to the pixel-family package.

Release a new version of pixel-family to GitHub and Typst Universe.

Use when triaging, grouping, or organizing GitHub issues by topic similarity — e.g., "group the issues", "triage issues", "organize issues into topics", "what issues are related", or any request to classify and consolidate open issues for easier collaboration. Also use when the user wants to file a summary issue that categorizes existing issues into named groups.

Use when an academic researcher wants to automate a workflow with a Claude Code skill, learn how skills work, or identify which of their repetitive workflows would benefit from a skill — guides discovery, selection, and hands-on skill creation

Use when adding or changing a zlp CLI command

Use before finishing changes to choose and run the right verification commands

Use when reviewing a paper by listening — reads LaTeX paragraphs aloud via macOS TTS, pauses after each to flag confusion/gaps as a naive reader, proposes fix options, applies edits inline, then advances. Invoke with /paper-read-aloud or /paper-read-aloud <section-number>

Use when self-reviewing a paper's writing quality section by section — dispatches a context-free subagent as naive reader or expert reviewer, identifies confusion, wordiness, and digressions, then presents fix options for non-mechanical issues