Arcanum

Use when deciding whether work merits a governed multi-agent dispatch and, when it does, proposing, tension-checking, confirming, registering, running, closing, and observing that dispatch through repository-local bindings.

Use when deciding whether work merits a governed multi-agent dispatch and, when it does, proposing, tension-checking, confirming, registering, running, closing, and observing that dispatch through repository-local bindings.

The design principle behind Principle 5 (pairwise tension) of the subagents-strategy constitution. Use when designing the angle set for an n≥2 subject group (role investigate or evaluate) so the agents' micro-vectors are structurally opposed and their correlated bias cancels under composition — not merely non-overlapping. This is the design/theory layer; check-tension is the executable gate that enforces it before the human confirm.

Use when: building a compact, task-ready context pack from selector-level evidence and obligation-linked excerpts.

Route repository work through installed Arcanum capabilities and execute validated capability-bound dispatches through host-native operations.

Route repository work through installed Arcanum capabilities and execute validated capability-bound dispatches through host-native operations.

Use when: validating or designing an Arcanum dispatch route that chains sigils, spells, owner capabilities, handoffs, gates, observability, techniques, and optional boundary/evidence contracts.

Route repository work through installed Arcanum capabilities and execute validated capability-bound dispatches through host-native operations.

Use when: executing or resuming the nearest evidence-backed bounded work-pack task or SWU end to end with explicit trade-offs, context building, gate checks, validation, owner-joined closeout synchronization, optional runtime handoff, and an optional non-closeout continuation route.

Use when: resolving blocker-level multi-option decisions before planning, implementation, document mutation, or other consequential changes continue.

Use when: installing or maintaining a repository-local compiled knowledge inventory that lets agents ingest once, index, tag, query, lint, and reuse context across sigils.

Use when: a terminal capability receipt names or implies a next owner route and the caller needs ranked probable routes, explicit authorization checks, one bounded dispatch, and a returned owner receipt without re-running the blocked source capability.

Use when: starting or operating a local Craft project ledger for recursive development contexts, blockers, enablers, decisions, gaps, definitions, next moves, validation, and recomposition.

Use when: polling the ide-extension manual session bridge, claiming and executing one ready request in the current session, or submitting a claimant-bound result or interruption.

Use when: starting or checking the ide-extension local workbench, making a request available for open-poll bridge testing, or running a bounded local bridge smoke.

Author or evolve an invocable Claude Code skill (SKILL.md) through a guided interview — gate whether it should be a skill, draft the routing description, set naming and frontmatter, write a dense body, self-check and verify it fires. Use when asked to create, scaffold, write, design, or edit a skill or slash-command.

Use when: maintaining canonical definitions, synchronized indexes, explanatory intuition, and downstream drift checks for critical terms.

Init-time anti-bias gate. Before the human confirm, two independent agents verify a proposed dispatch sheet is genuinely tensioned against the rubric this skill owns — the four anti-bias tests plus a cross-group coherence check; the sheet reaches the human only if BOTH pass. Any reproval or disagreement returns it to the strategist. Gate infrastructure: not itself a dispatch, not subject to its own gate.

Subagent dispatch that PROPOSES and pre-registers a falsifiable experiment — the designer authors a success/failure criterion fixed BEFORE any run and the skeptic validity-checks it, producing a frozen criterion.md. Running the probe and adjudicating survived-vs-falsified is a SEPARATE downstream step, not part of this dispatch. Routed here from domainspec-subagents-strategy as the LIVE type skill for `dispatch_type: experiment`. Defines experiment-type judgment only — the pre-registered criterion, the design-time validity gates, and the SURVIVED/FALSIFIED/INVALID verdict vocabulary (SURVIVED/FALSIFIED rendered later, at run). Use research when the question is whether a NEW claim survives by coverage; use review to attack an EXISTING artifact; use experiment when you are pre-registering a probe against a criterion fixed BEFORE running.

Operating guide for running a `dispatch_type: research` subagent dispatch: shape it, find the evidence, run it, gate it, and write research.md + findings.md. Routed here from domainspec-subagents-strategy.

Red-team dispatch over EXISTING artifacts (skills, constitutions, code, docs) — attack what exists and hunt flaws to improve it. Routed here from domainspec-subagents-strategy as the LIVE type skill for `dispatch_type: review`. Defines review-type judgment only — attack lenses, severity taxonomy, verification discipline, and the change-request findings shape. Use research (not review) when the question is whether a NEW claim/candidate survives; use review when the target already exists and the deliverable is verified change requests.

Run a multi-agent parallel investigation that maps cross-layer tensions before any fix — decompose by concern, explore in parallel, synthesize contradictions, gate on a human.

Operate the ui-prototyping-studio as Claude Code — generate UI variants and drive the explore/annotate/mutate loop through the studio CLI. Every change is reversible; the human owns the one irreversible edge (export). Use when a human asks to explore, annotate, or revise UI prototypes in this project.

Close a session and create a vault conversation node

Use when: mapping, distilling, validating, or evolving a governed knowledge vault with roles, confidence, premises, sessions, edge rules, branch-aware ontology, and ontology conventions.

Use when: turning user-supplied context about an object, artifact, architecture, codebase, process, or system into a visual layered HTML explanation page with properties, components, flows, dependencies, and evidence boundaries.

Use when: turning a UI iteration session into saved lessons and reusable ux-patterns that later feed ux-evidence-validator and ui-prototyping-studio.

Use when: starting or operating a link-indexed local Craft project ledger for recursive development contexts, blockers, enablers, decisions, gaps, definitions, next moves, pending-by-node status, validation, and recomposition.

`goal` drives a Craft-backed work graph through an autonomous, fail-closed control loop. It reads the current frontier, classifies each node by risk, routes work to the owning capability, gates closure with audit evidence, stages ledger deltas, and promotes only behind an explicit approval token.

Use when: formalizing, routing, validating, promoting, or retiring an Arcanum discipline so a recurring cross-capability practice gets a card, catalog entry, and a hardening route without claiming sigil or spell authority.

Use when: claiming one GitHub Project issue, refining its context, invoking define/design/plan as needed, executing through Task Session, and opening a linked PR with verification evidence.

Use when routing a verification target to status-qualified owner lanes and joining native evidence receipts into a VERIFICATION-WEAVE parent receipt without promoting owner outputs.

Use when: turning a vague refinement target, design concern, folder, idea, repository area, or existing work-pack into a refined seed, design, or non-executed plan through the Arcanum refinement loop.

Use when selecting and running invoke template validation prompts in Codex, including picking a task by ID, template, complexity, or the next unrun prompt and saving the Invoke Result output.

Use when converting an Arcanum work-pack task, SWU, or explicitly selected one-shot stream into a compact native Codex /goal command with sidecar context, profile policy, verification surface, constraints, boundaries, iteration policy, and blocked stop condition.

Use when: formalizing, routing, validating, promoting, or retiring an Arcanum discipline so a recurring cross-capability practice gets a card, catalog entry, and a hardening route without claiming sigil or spell authority.

Use when: starting or operating a link-indexed local Craft project ledger for recursive development contexts, blockers, enablers, decisions, gaps, definitions, next moves, pending-by-node status, validation, and recomposition.

Run a 5-dimension expert design review on any HTML artifact in the project — Philosophy / Visual hierarchy / Detail / Functionality / Innovation, each scored 0–10. Outputs a single self-contained HTML report with a radar chart, evidence-backed scores, and three lists: Keep / Fix / Quick-wins. Use when the brief asks for a "design review", "design critique", "5 维度评审", "design audit", or "what's wrong with my design".

Admin / analytics dashboard in a single HTML file. Fixed left sidebar, top bar with user/search, main grid of KPI cards and one or two charts. Use when the brief asks for a "dashboard", "admin", "analytics", or "control panel" screen.

Parse a structured design brief written in I-Lang protocol format into a concrete design spec. Eliminates ambiguity from vague requests like "make it professional" by requiring explicit dimensions: palette, typography, layout, mood, density, and constraints. Trigger keywords: "design brief", "create a design brief", "ilang brief", "structured brief".