skills

Order user-facing questions by dependency. Use when two or more clarifications may depend on one another, a high-impact fork should be resolved before details, a plan needs a decision-tree interview, or a visual choice should be previewed before asking.

Critically evaluate a planned or existing feature against the project's vision and decide whether to keep, sharpen, park, or cut it — favoring depth on the core loop over breadth of surface area. Use for "is this in scope", "does this fit the vision", "prune scope", "feature creep", "scope creep", "depth not breadth", "should we remove this", "audit features against the vision", "why does this feature exist", "agents keep adding WOW features", "keep this out of scope", maintaining an OUT_OF_SCOPE.md ledger, patrolling a repo for re-added parked/cut features, or gating proposed work items (beads/plans) before they enter the backlog. Read-only by default: it recommends and records; it does not delete code without explicit approval.

Run Codex in a persistent tmux session with signal-based completion so long tasks survive conversation death. Use when work may take 5+ minutes, needs background execution, or must remain retrievable after the calling session ends.

Deploy, rollback, debug, and operate Docker or VPS stacks, Pages or edge frontends, packages, and app-store releases using a skillbox client overlay. Also design or migrate repo-owned local self-test and self-release paths that move avoidable build, test, and deploy work off paid GitHub Actions. Use when deploying code, reducing Actions spend, creating a self-deploy or self-test path, checking health, reviewing logs, comparing env rollout plans, validating package publishing, or working in SSH dev environments where production constraints matter.

Decompose complex work into a Beads-backed execution graph with generated `WORKGRAPH.md` views, then run an NTM-style swarm by ready frontier with no write overlap. Use before parallel execution for large-ish, UI-facing, multi-file, multi-domain, naturally parallel, naturally orchestrated, subgoal-oriented, or review-sensitive tasks.

Plan new multi-repo domain slices, assess plan quality, or orchestrate implementation from an accepted slice plan. Use for "plan the X slice", multi-repo feature planning, API contract design, "implement the X slice", or slice-quality review; not for bug fixes, small refactors, or single-repo work. When a slice premise depends on live external reality rather than internal repo evidence, run a bounded GPT-5 Pro / Deep Research gate before committing to the plan.

Three modes: audit active implementations against plan, retire completed slices into lean COMPLETED.md summaries, and retire-session consolidation of DONE session plans. Use for review/audit slice, what's left checks, retire/close out/finalize slice, or retire session plans workflows.

Friction-removal retrospective for agent sessions. Use when the user says "lube", "$lube", "/lube", "friction", "several frictions were observed", "how do we unblock this", "avoid this in the future", or asks to prevent similar or adjacent blockers across future sessions.

Review and score your AI prompting quality. Analyzes Claude Code and Codex conversation history to evaluate clarity, context, collaboration, and outcomes on a 23-point scale. Tracks scores over time for week-over-week progress. Use when asked to "review my prompts", "score my prompts", "benchmark myself", "rate my prompting", "analyze my conversations", "prompt quality check", "show my trend", "prompt progress", "backfill my history", "backfill next", "list weeks", or "/prompt-reviewer". Also triggers on "how am I doing as a prompter?" or "what can I improve?" or "show my prompt history" or "what weeks haven't I reviewed?".

Identify the single most accretive, highest-leverage next move, then by default define a goal sequence that encompasses it, turn it into a set of no-ragrets beads, and pursue the goal. Use when tagged with /smart, when the user wants to identify the highest-leverage next move, most impactful addition, or smartest thing to do next, when the user asks for smart goal mode, ambitious goal tracker, Gantt or ganntt plan, or when the user wants /smart to continue, iterate, or loop through bounded rounds until a concrete goal or success criterion is met.

Pull user-reported issues from the SPAPS issue reporting platform and route each one to the right sibling skill via a per-client overlay registry. Use when asked to "pull spaps feedback", "check user issues", "what's in the feedback inbox", "triage spaps issues", "find issues about X", or "which skill should handle this issue".

Choose, author, persist, edit, stack, and open Mermaid/MMDX diagrams in Buildooor. Use when the user invokes /mmdx or $mmdx, asks to create or update a private/owned Buildooor MMDX diagram, asks for the best .mmd/.mmdx diagram type for a context, wants a Mermaid diagram created from prose, needs a mermaid.live/edit pako URL, asks to preview a .mmd file, asks for chart drilldown/stacking, or wants to decode/check Mermaid Live URL fragments.

Create, update, review, and package skills for AI coding agents. Also manages client overlays (create, list, validate, match, migrate). Use when asked to "create a skill", "make a skill", "new skill", "skill template", "design a skill", "build a skill", "review this skill", "improve this skill based on past runs", "when did we last use this skill", or when working with SKILL.md files, frontmatter, bundled resources (scripts/, references/, assets/), .skill packaging, or Claude/Codex transcript-driven skill reliability. Also triggers on "how do I make a skill", "skill best practices", "skill structure", "skill reliability", "operator evidence", "create an overlay", "list overlays", "check my overlays", "which overlay matches", "migrate overlays", or requests to extend an agent's capabilities with reusable workflows.

Audit MMDX registry surfaces including INDEX.mmdx freshness, repo-local tracker placement, diagram stack links, Mermaid preflight validity, stale diagram drift, and plan-to-diagram evidence freshness. Use when placement or registry decisions depend on whether MMDX files are indexed, valid, linked, current, or stored in the right repo, plan root, skill template, or Skillbox-config overlay.

Create and update repo-local linked MMDX "where are we" project status stacks, then publish/edit the project's existing status diagram in the operator's `buildooor` mmdx account. Use when the user asks for "$mmdx $eli-me" plus "$reality-check-for-project", "where are we", "where we're at", primary user flows, epics, gantt/status, ERD inclusion, deep-linked project maps, "update the project status mmdx", "push it to my mmdx account", or a durable visual reality-check artifact stored inside the current project.

Produce copy-pasteable mega-prompts for external deep research tools (ChatGPT Deep Research, Perplexity Deep Research, Claude Research), Oracle-ready prompt handoffs when the current run should execute GPT-5 Pro / Deep Research directly, and high-detail image-creation prompts for ChatGPT image creation via Oracle. Use when the user asks for "a prompt for another agent to research X", "mega prompt for deep research", "draft a research prompt", "make a prompt to paste into ChatGPT deep research", "prompt for another agent to do all the Y", "image prompt", "make an image prompt", "use this image/style/contact sheet as visual inspiration", "describe these references for an image generator", "get this off the page", "better vibes", "image to the left/right", or when another skill needs a bounded external-reality pass before a strategic decision or document update. Not for inline research the current agent can do with WebFetch/WebSearch, and not for prompts asking another agent to write code or edit files.

Decide whether the current run needs a bounded external-reality pass before committing to a strategic call, then route it to web-check, deep-research-prompt, thesis-gtm, research-paper, or skip. Centralizes the escalation gate that thesis-gtm, power-map, build-vs-clone, domain-planner, wiki-forge, and readme-writing each restated. Use for "escalate", "/escalate", "do we need a Pro pass here", "should we Oracle this", "is this gated by external reality", "bounded Deep Research pass", or when a calling skill's plan/recommendation depends on facts outside the local repo and wiki.

Create or update a private eli-me skill that captures a user's preferred explanation style. Use when the user says "eli-me-maker", asks to make an "eli-me" skill, wants a reusable personalized communication-preference skill, or wants examples-based questions that discover how they like concepts defined, explained, scoped, and checked.

Commit working changes with clean, logical batches, including from a fresh repo context where no prior ownership is established. Use for "commit", "checkpoint", "commit everything", "commit my changes", "save progress", "/commit", "initial cleanup commit", or when a dirty tree needs batching, .gitignore decisions, and OSS/privacy-aware staging.

Assess a user's local environment (repos, tools, skills, Claude config), generate a skillbox client overlay, and provision a working skillbox container on DigitalOcean or locally. Use for "up up up", "set up my skillbox", "quickstart", "new box", "onboard me to skillbox", "spin up a box", or when a user wants to go from zero to a working skillbox environment.

Compute CRAP-style hotspot scores across Rust, Python, TypeScript, and Swift, print a ranked risk report, end with a FINAL_SCORE line, and suggest hardening follow-ons. Use for "/crap", CRAP score, risk score, hotspot ranking, weak coverage, or taking a scoped score under a threshold.

Run scoped mutation testing with the right adapter, keep the baseline test path green, triage surviving mutants, and feed the results into a hardening loop. Use for "/mutate", mutation tests, surviving mutants, weak tests, or post-CRAP hotspot hardening.

Scaffold backend or frontend domain code from an existing slice plan using a shared stack-aware workflow. Use for "scaffold a domain slice", "implement the backend for {slice}", "implement the frontend for {slice}", or "scaffold {slice}" after domain-planner finishes a slice plan. This is the only supported domain scaffolder skill.

Server connection reference and targeted live status checks for containerized environments. Use when asking "ssh info", "server status", "what's running", "container status", "how do I connect", "check production", "which containers", or "show me the health checks" and the answer depends on a skillbox client overlay with host and service details.

Build, modify, or review agent-facing CLIs for compact output, strong defaults, and clean shell UX. Use when a tool is mainly consumed through shell execution by Claude Code, Codex, Cursor, or similar agents.

Audit skill registry surfaces including skill-repos.yaml, SKILL.md manifests, Skillbox runtime manifests, SBP policy, MCP visibility, default bundles, overlays, and private/public placement. Use when placement or adoption depends on whether a skill, bundle, overlay, or MCP server is registered, visible, fresh, scoped correctly, or owned by opensource/skills, skillbox, skillbox-config, or a private skill corpus.

Design and implement data-intensive shadcn/ui tables that stay scannable at scale. Produces a column-priority plan, a filter/sort contract, density and state specs, and a TanStack Table v8 + shadcn implementation grounded in UX psychology (Hick, Fitts, F-pattern, recognition-over-recall) and Tufte-style data density. Use when asked to "design a data table", "build a shadcn table", "make this table easier to use", "pick columns for this table", "add filtering/sorting", "tame a table with too many columns", "make this table less overwhelming", or when a dense list/grid/report UI needs a real IA pass instead of dumping every field into columns.

Best practices for Remotion - Video creation in React. Use when working with Remotion compositions, animations, SVG-first motion architecture, stack selection, or video rendering. Covers project setup for a shared Remotion hub, animation patterns, timing/interpolation, motion-spec design, preview/runtime boundaries, audio, captions, and media handling.

Transform social and search trends into research, PSEO pages, videos, and copy. Use for "what's trending", "content ideas", "viral content", PSEO, trend research, video content ideas, or planning what to create next, with Research, PSEO, and Video modes.

Apply non-deterministic DRY refactoring to non-code artifacts — wiki concept pages and skill bundles. Identify candidates to inline, extract, compose, unify, split, or promote so the corpus stays modular without restating the same idea in five places. Use for "wiki dry", "/wiki-dry", "dry the wiki", "dry my skills", "which skills should be sub-skills", "merge these concepts", "split this concept", "find duplicate concepts", "concept refactor", "skill consolidation", "abstract these up", or when a wiki vault or skill collection feels redundant, sprawling, or sub-optimally organized.

Run /dueling-idea-wizards on a project or topic after first grounding the agents in the wiki. Thin wrapper that calls /wiki query to build a shared prior, then launches /dueling-idea-wizards with that prior baked into the study phase. Use for "wiki duel", "grounded duel", "duel with wiki context", "duel but check the wiki first", or when an adversarial brainstorm should start from accumulated wiki knowledge rather than a cold read of the repo.

Identify the highest-lever concept in a wiki vault, stress-test it adversarially via multi-model duel, optionally validate the final synthesis against current external reality with GPT-5 Pro + Deep Research, and file it back. Use for "wiki forge", "wiki synthesize", "deepen the wiki", "stress test the wiki", "find the highest lever", "forge a concept", "adversarial wiki synthesis", or when the wiki needs to confront its own assumptions.

Detect UI drift across a frontend codebase — ad-hoc color/spacing/typography values vs design tokens, duplicate component variants, Tailwind arbitrary values, className soup, orphan primitives — then produce a ranked consolidation plan. When variants are consolidated, the old code is moved to a numbered per-family archive (single-page index) so no archetype is lost. Use when the user asks about "UI drift", "inconsistent UI", "poorly modularized UI", "too many button variants", "clean up the design system", or "consolidate components across projects".

Generate dense research-paper-style pages on any topic plus companion X and LinkedIn drafts that keep the same thesis. Also has a discover mode that mines recent session activity (via cass) and existing paper coverage to propose accretive topics rooted in actual work. Use for "research paper", "write a paper on", "research page", "/research-paper", "discover topics", "what should I write about", "find me something I'd like", "find a strong angle", "surprise me with a research topic", or internal write-ups on a topic, especially when the paper should be grounded in existing wiki concepts, pressure-tested with wiki duels when needed, externally validated with GPT-5 Pro / Deep Research, and then fed back into the wiki as a new source note.

Maintain a Karpathy-style LLM Wiki in an Obsidian vault. Default bare invocation ("wiki", "/wiki", "$wiki", or a wiki tag with no operation) runs orient mode: reality-check the vault, choose one smart wiki goal, and get to work on the safest highest-leverage wiki operation. Also use for "wiki ingest", "wiki query", "wiki lint", "ingest this source", "what does the wiki say about", "lint the wiki", or when connecting knowledge across repos and skills.

Decide whether work should live in an existing repo, be extracted upward into a shared home, adopt existing open source, borrow patterns, or be built from scratch. Use for "where should this go", "should this be extracted", "should this become a skill/shared package/helper repo", "is there an open source repo for this", "am I recreating the wheel", "should we fork/build this", "audit this codebase for build-vs-buy violations", "what can I get from this repo", "mine this repo", "prospect this", or when a plan should consider both ecosystem fit and the user's current repo portfolio, including dogfood buckets plus registry-backed placement checks for skill, SBP, MCP, or MMDX evidence freshness. When the blocker is current external reality rather than local code inspection, use `escalate` before finalizing the recommendation.

Audit a repository's public docs for drift against the active codebase, grade OSS readiness, and produce a ranked cleanup queue. Use when asked to "audit docs", "check OSS readiness", "find stale docs", "grade OSS readiness", "what docs are wrong", or "prepare this repo for open source".

Harden SwiftUI iOS surfaces by extracting a shared visual language from the strongest existing screen, reusing it across auth/onboarding/home flows, and pairing the UI pass with a repo-owned analytics contract plus fixture-backed validation. Use when an iOS app has one polished surface and weaker sign-in, onboarding, settings, or companion screens; when someone asks to "make onboarding look like the main app", "standardize the design system", "harden SwiftUI auth flow", or "add one analytics contract for iOS".

Decide when an agent should recommend booking the human operator for expert help, then prepare a concise booking handoff. Use when a user asks to "book me", "hire me", "bring in the operator", "get expert help", "schedule a consult", or when a task crosses into high-stakes architecture, production, security, payment, compliance, multi-repo strategy, repeated agent churn, or ambiguous delivery risk that needs a human expert.

Build persuasion-forward charts and dashboard components, defaulting to shadcn/ui plus Recharts for React projects. Use when the user says "chart crimes", "make this chart persuasive", "visualize this to prove our point", "choose the best chart for this narrative", asks for a shadcn chart, or wants matplotlib, Plotly, seaborn, or Recharts code that emphasizes a claim while keeping the underlying numbers accurate and disclosed.