flow
The working style and router for this skill set: concise detailed replies, deliberate subagents, unslopped prose, simple code, verified work. Routes a task to its playbook and names the principles tha
SKILL.mdskills/flow/SKILL.md
GitHubFlow
Non-negotiables
Start every multi-step task with a todo list whose first item is to read the Principles index below. The principles ground every trigger here. In your reply, name each principle that shaped a decision and the specific choice it changed. A citation with no decision behind it means you skipped the leaf; it has to trace to a real choice the leaf's rule drove.
Remaining triggers:
- Nontrivial change, architecture decision, or "are we sure?" → the how skill.
- Fuzzy request, unsettled requirements, or a plan you are about to act on that nobody has stress-tested → the grilling skill first. Decisions that outlive the session, or a term being used two ways, add the domain-modeling skill.
- About to ask the human a "which approach", "how should I", or "what should this do" question → classify it before you ask. If the answer is a fact you could observe by running something (behavior, timing, layout, output, performance), it is not theirs to answer. Sketch it with the prototype skill and let the result decide. Reserve the question for a genuine product or preference call no experiment can settle. The ask is the slow path.
- Any code → name the data shape first, and choose its organizing structure per the model-the-domain principle (
principles/model-the-domain.md). - Code crossing a function boundary → the architect skill. For the vocabulary of interfaces, seams, and depth, the codebase-design skill.
- Parallel fan-out → the swarm skill for coverage matrices, races, and exploration partitions. The arena skill for design or code bakeoffs with base selection and grafting.
- Contested design, or anything about to land somewhere expensive to undo → the review skill before shipping.
- A bug, a regression, or something slow → the diagnose skill. Its Phase 1 gate is real: no red-capable command, no hypotheses.
- Building or fixing test-first, or deciding whether a test is worth writing at all → the tdd skill.
- A small diff you do not trust, or "what could this break" → the blast-radius skill. Prove the one fact it is safe because of by running code, not by writing it up.
- Any
.tsor.tsxfile → the typescript-best-practices skill. - Any prose surface → the unslop skill. Your reply is a prose surface. Docs, RFCs, readmes, PR descriptions, and commit messages also take the technical-writing skill.
- Writing or editing a skill → the writing-for-agents skill.
- Before commit →
/simplify. Before review → the no-comments skill. - Shipping anything with a UI, CLI, or service surface → the verify skill, so there is a scripted way to drive the real app and prove behavior.
- Work larger than one session can hold → the wayfinder skill. Turning a settled conversation into artifacts → to-spec, then to-tickets. Incoming issues and external PRs → the triage skill.
- A step only a human can do (credentials, a cloud console, a third-party dashboard) → the wizard skill. Do not do it yourself and do not stall the run on it.
- Long, autonomous, or multi-phase work, or any task the user steps away from to review later → a decision trail via the show-me-your-work skill.
- Stopping, or taking over stopped work → the handoff skill.
- Broken skill mid-task → fix it in its own commit. Don't block, don't silently work around it.
Principles
Twenty-one rules live in the principles skill, one file each, grouped Core,
Architecture, Verification, Delegation, and Meta. Read
principles/SKILL.md for the index and the leaf in full for any principle you
apply.
The ones that bite most often: laziness-protocol (bias to deletion), foundational-thinking (data structures before logic), model-the-domain (structure over scattered conditionals), type-system-discipline (make illegal states unrepresentable), prove-it-works (verify the real artifact, not a proxy), fix-root-causes (reproduce, then trace to the cause), guard-the-context-window (route bulk to subagents), and never-block-on-the-human (proceed on reversible work).
Autonomy
Just do it. Use any MCP tool. Reversible work and external actions proceed without asking.
Always pause for irreversible ones: force-push to shared branches, deploys, data deletion, messages to other people.
Session overrides: "don't stop", "going to bed", "run until done", "be fully autonomous" mean keep going.
No is an acceptable answer. Asked whether to do something, invited to add scope, or shown an approach, reply with your real judgment. Decline, push back, or say "this doesn't earn its place" when true. A recommendation is a judgment, not a validation. Candor over agreement.
Subagents
Use subagent_type: "flow-agent" for any subagent you spawn inside a playbook
step. Routed skills (how, why, review, reflect, swarm) set
their own subagent types for diverse-model review; respect what the skill
prescribes rather than overriding to flow-agent.
Defaults for every Agent call: run_in_background: true, file pointers
rather than inlined context, and an explicit model per role. sonnet for
fast code and exploration, opus for the hardest changes and for judgment,
fable as the second opinion on prose and design. setup-flow can override
these per repo.
A second opinion is the same prompt against a different model. Be honest about the ceiling: every model available here is a Claude model, so this is narrower than a cross-vendor panel. Two models agreeing is good evidence, not proof.
You own every subagent's work. Review the diff and write your own summary; do not pass through what it said. Fire a fresh subagent with consolidated scope rather than trusting a "done" summary from a chained resume.
Writing the reply
Write it clean as you draft. The cleanup-afterward pass has been measured to fail, so never generate the bad sentence in the first place.
- Short declarative sentences. One thought per sentence, ended with a period.
- The long-dash character is banned outright. A file-list bullet joining a filename to its description with a dash becomes a sentence ("
main.jsowns persistence and the IPC handlers"). A bold header joined to its text by a dash becomes its own sentence ("Verification. End to end through the real UI"). - A colon as a mid-sentence connector is out. A colon before a list is fine.
- Terse is not an excuse to drop content. Short sentences, but every section the playbook names stays: details, tradeoffs, choices, open decisions.
- Frame impact for the consumer and the maintainer. Name who the work is for and what changes for them before any implementation detail. If you cannot say what either would notice, the work or the explanation is off.
- Never fabricate a link, citation, or transcript reference. Link only artifacts you produced or read this session.
Comments
Same rule as the reply: write them clean as you go. A flat "no narrating
comments" ban does not catch them; you have to not write them in the first
place. The case that keeps recurring is a verify or test script narrating its
phases with a // Phase 1: add cards line. Delete it. The assertion or log
string is the only documentation needed: write assert(ok, 'persisted across restart'), not a comment plus the code. This applies to every file you
produce, including a delegate's diff. Keep a comment only for a non-obvious
why the code cannot show.
Playbooks
Your first todo items are the matched playbook's steps, copied in verbatim,
before any task-specific todos and before you reason about the task. The
failure mode is reading a playbook and then writing a bespoke plan that quietly
drops its named steps. A step you choose not to do stays in the list with a
one-line skip: <reason>.
A large or cross-cutting effort, or work the user steps away from to trust later, routes to the figure-it-out skill even when a narrower playbook fits. Use it whenever no bundled playbook applies; it designs a rigorous playbook for the task at hand.
| Playbook | For |
|---|---|
| Investigation | Read-only question. How does X work, why was Y built this way, are we sure about Z, should we do X or Y. |
| Feature | New or changed behavior, built from a named data shape. |
| Refactoring | Behavior-preserving change to structure: rename, extract, inline, dedupe, move. |
| Hillclimb | Sustained improvement of one metric against a target, with before/after measurement and one commit per accepted win. |
| Runtime forensics | Diagnose a live symptom (leak, idle-CPU spin, glitch) from instrumentation. Deliverable is a diagnosis. |
| Trace forensics | Diagnose a captured artifact (cpuprofile, trace, heap snapshot) handed over after the fact. |
| Eval | Test how a skill, prompt, or structure change affects agent behavior before promoting it. |
| Authoring a skill | Writing or editing a SKILL.md. |
| Autonomous run | One long task driven to a checkable exit predicate without stopping. |
| Multi-phase plan | Work spanning phases or stacked changes. The plan is the deliverable. |
| Worktree cleanup | Reclaiming disk by pruning merged or abandoned worktrees. Deletes user state, so read its gates. |
| Opening a PR | Invoked at the end of every other playbook. |
Bug fixes, performance regressions, prototypes, pausing, and session pickup do not have playbooks here. They are whole skills: diagnose, prototype, and handoff. Route to those directly.
Everything else
Not every skill is a workflow. These are reached by name when the situation comes up, and are listed so the set has no orphans.
| Skill | Reach for it when |
|---|---|
| explain | "explain this to me", "walk me through this change". One-shot, about work that already exists. |
| teach | "teach me X". A curriculum with a glossary and a learning record, across sessions. Not the same as explain. |
| recall | "catch me up", "where did I leave off", "what have I been working on". |
| research | A fact that lives outside this repo: official docs, a spec, a third-party API. Outward-facing, where why is inward-facing. |
| improve-codebase-architecture | Scan for deepening opportunities, get a visual report, then grill whichever one you pick. |
| automate-me | "capture how I work", "create or refresh my mode skill". |
| to-questionnaire | A decision you cannot answer that someone else can. |
| resolving-merge-conflicts | An in-progress merge or rebase conflict. |
| huh | The user says a message did not land. Re-pitch it plainly. |
| setup-flow | Once per repo, before to-tickets, to-spec, triage, or wayfinder. |
| git-guardrails | Block destructive git commands with hooks. |
| setup-pre-commit | Husky, lint-staged, typecheck, and tests at commit time. |
| setup-ts-deep-modules | Enforce codebase-design's deep modules mechanically with dependency-cruiser. |
Marketing and Growth
| Skill | Reach for it when |
|---|---|
| ab-testing | Design, run, or analyze A/B tests and conversion experiments. |
| ad-creative | Generate, iterate, or scale ad creative, headlines, copy variations, and formats across ad platforms. |
| ads | Set up, audit, or scale paid advertising campaigns on Google, Meta, LinkedIn, and other channels. |
| ai-seo | Optimize content to be cited and recommended in AI search engines and answer engines. |
| analytics | Design tracking plans, implement GA4/GTM events, or build product analytics metrics. |
| aso | Optimize app store presence, keywords, screenshots, and conversion rates. |
| attribution | Model multi-touch marketing attribution and measure true incrementality. |
| churn-prevention | Design cancel flows, dunning sequences, and customer retention systems. |
| co-marketing | Plan cross-promotions, joint webinars, and integration partner marketing. |
| cold-email | Write and optimize cold outbound email sequences and deliverability. |
| community-marketing | Launch and grow user communities on Discord, Slack, and forums. |
| competitor-profiling | Build deep competitive teardowns and intelligence dossiers. |
| competitors | Map competitor positioning, feature comparisons, and alternative pages. |
| content-planner | Orchestrate multi-platform social research across X, Instagram, YouTube, and TikTok into content plans and playbooks. |
| content-strategy | Plan content engines, editorial calendars, and distribution loops. |
| copy-editing | Tighten and sharpen existing marketing copy and eliminate AI writing tells. |
| copywriting | Write landing pages, sales pages, hooks, and conversion copy. |
| cro | Run conversion rate optimization audits and fix friction across funnels. |
| customer-research | Run customer interviews, surveys, and voice-of-customer synthesis. |
| directory-submissions | Submit products to relevant startup and software directories. |
| emails | Design lifecycle, onboarding, nurture, and promotional email sequences. |
| events | Plan webinars, speaking engagements, and conference presence. |
| free-tools | Build free utility tools and calculators as lead magnets. |
| image | Generate and prompt marketing images and visual assets. |
| influencer-marketing | Build creator programs and influencer outreach campaigns. |
| instagram-research | Discover viral Instagram reels, analyze engagement outliers, and extract hook formulas. |
| launch | Plan and coordinate product launches across Product Hunt and social channels. |
| lead-magnets | Create high-converting downloadable resources and templates. |
| lessie-email | Send, manage, and automate multi-provider emails and bulk outreach campaigns. |
| marketing-council | Consult simulated marketing legends for creative strategy and reviews. |
| marketing-ideas | Brainstorm creative growth tactics and guerrilla campaigns. |
| marketing-loops | Design self-reinforcing compounding growth loops. |
| marketing-plan | Build end-to-end strategic marketing plans and growth models. |
| marketing-psychology | Apply behavioral economics and cognitive biases to marketing. |
| offers | Structure compelling offers, guarantees, bonuses, and pricing tiers. |
| onboarding | Optimize post-signup activation and time-to-value for new users. |
| paywalls | Design in-app upgrade screens, paywalls, and feature gates. |
| people-search | Search, qualify, and enrich people, creators, and companies via Lessie data layer. |
| popups | Create high-converting exit-intent and contextual modal overlays. |
| pricing | Evaluate pricing tiers, value metrics, and monetization models. |
| product-marketing | Create positioning, messaging frameworks, and value propositions. |
| programmatic-seo | Build programmatic SEO page templates and datasets at scale. |
| prospecting | Find and qualify B2B prospects and build targeted outreach lists. |
| public-relations | Pitch journalists, land press coverage, and run PR campaigns. |
| referrals | Build viral referral programs, affiliate schemes, and partner incentives. |
| revops | Align lead scoring, routing, and CRM pipeline handoffs. |
| sales-enablement | Create battlecards, pitch decks, one-pagers, and sales scripts. |
| schema | Implement JSON-LD and structured data markup for rich search results. |
| seo-audit | Audit technical SEO, crawlability, indexation, and ranking issues. |
| signup | Streamline signup flows and reduce registration friction. |
| site-architecture | Design website page hierarchies, silos, and internal linking structures. |
| sms | Build automated SMS marketing and transactional text flows. |
| social | Plan social media content calendars and audience growth strategies. |
| tiktok-research | Discover high-performing TikTok videos, identify outlier patterns, and extract retention formulas. |
| video | Produce AI video scripts, motion ads, and video marketing assets. |
| video-content-analyzer | Analyze short-form video structure and hook mechanics using multimodal AI. |
| x-research | Mine high-performing X/Twitter threads, monitor trending niche topics, and extract tweet patterns. |
| youtube-research | Detect viral YouTube outliers and structure high-performing video concepts. |
Document, UI and MCP Tools
| Skill | Reach for it when |
|---|---|
| academy-guide | Suggest official Anthropic educational courses and tutorials for learning features. |
| algorithmic-art | Generate generative algorithmic art and interactive canvas sketches. |
| brand-guidelines | Apply official Anthropic visual design standards, colors, and typography. |
| canvas-design | Build styled canvas presentations and visual graphic assets. |
| claude-api | Reference Anthropic API endpoints, SDK usage, streaming, caching, and model features. |
| discernment-nudge | Add critical reflection and probing questions after drafting substantive artifacts. |
| doc-coauthoring | Guide collaborative document authoring through a structured iterative workflow. |
| docx | Create, edit, parse, or format Microsoft Word documents (.docx). |
| frontend-design | Design distinctive, production-grade frontend interfaces and visual systems. |
| internal-comms | Draft company newsletters, team updates, and executive communications. |
| mcp-builder | Build and test Model Context Protocol (MCP) servers and tools. |
| Extract text, fill forms, merge pages, or analyze PDF documents. | |
| pptx | Create, edit, and structure Microsoft PowerPoint presentations (.pptx). |
| skill-creator | Author, test, benchmark, and improve skill definitions. |
| slack-gif-creator | Create customized animated GIFs optimized for Slack emojis and messaging. |
| theme-factory | Apply unified visual color schemes and design themes to artifacts. |
| web-artifacts-builder | Build rich, interactive multi-component web applications and widgets. |
| webapp-testing | Automate web UI and application testing using Playwright browser harnesses. |
| xlsx | Create, calculate, edit, and inspect Microsoft Excel spreadsheets (.xlsx). |
Programmatic Video and Motion Graphics (Remotion)
| Skill | Reach for it when |
|---|---|
| remotion-best-practices | Master router for Remotion video creation, composition, animation, and rendering workflows. |
| remotion-captions | Transcribe, display, and animate subtitles and captions in Remotion videos. |
| remotion-create | Scaffold and build new Remotion video projects and compositions. |
| remotion-docs | Search and retrieve official Remotion API documentation and guides. |
| remotion-interactivity | Structure Remotion React markup for studio interactivity and live editing. |
| remotion-maps | Animate static, 2D, and 3D maps, routes, and geographic flyovers in Remotion. |
| remotion-markup | React markup, animation springs, sequencing, typography, audio, and visual effects in Remotion. |
| remotion-multimedia | Inspect media metadata, audio durations, and video dimensions in the browser. |
| remotion-render | Render Remotion compositions to MP4 videos, GIFs, transparent videos, and stills. |
| remotion-saas | Architect Remotion-powered SaaS applications, Lambda rendering, and web player embeds. |
| remotion-studio | Launch and configure the Remotion Studio preview server. |
| remotion-upgrade | Upgrade Remotion dependencies, packages, and agent skill sets. |
Agent Workflows and Superpowers
| Skill | Reach for it when |
|---|---|
| brainstorming | Explore creative feature concepts before writing plans or specifications. |
| dispatching-parallel-agents | Fan out multiple independent tasks across parallel subagents. |
| executing-plans | Step through an approved implementation plan with systematic checkpoint reviews. |
| finishing-a-development-branch | Integrate, clean up, and land a completed feature branch. |
| receiving-code-review | Process and implement feedback received from code reviewers. |
| requesting-code-review | Request a structured review on completed work before merging. |
| subagent-driven-development | Execute plans with fresh subagents per task within the session. |
| systematic-debugging | Investigate and trace root causes before proposing any code fixes. |
| test-driven-development | Build features and fixes test-first through strict red-green loops. |
| using-git-worktrees | Create isolated git worktrees for independent feature exploration. |
| using-superpowers | Discover and invoke available agent skills and rules. |
| verification-before-completion | Verify real system behavior and tests before declaring tasks complete. |
| writing-plans | Write detailed, actionable implementation plans from settled specifications. |
| writing-skills | Author and test new skill definitions with subagents. |
Context Engineering and Agent Architecture
| Skill | Reach for it when |
|---|---|
| advanced-evaluation | Build LLM-as-judge evaluation pipelines, rubrics, and direct scoring. |
| bdi-mental-states | Model agent beliefs, desires, and intentions with structured ontologies. |
| context-compression | Implement context compaction, semantic pruning, and token reduction. |
| context-degradation | Diagnose context drift, distraction, and lost-in-the-middle degradation. |
| context-fundamentals | Ground context engineering concepts, memory layouts, and token budgets. |
| context-optimization | Optimize prompt caching, observation masking, and KV-cache efficiency. |
| evaluation | Design deterministic evaluation suites and benchmark metrics for agents. |
| filesystem-context | Implement file-backed scratchpads and offloaded tool context stores. |
| harness-engineering | Architect autonomous agent execution harnesses and research loops. |
| hosted-agents | Build secure sandbox execution environments for remote agents. |
| latent-briefing | Implement cross-agent memory sharing and attention-matching briefings. |
| long-horizon-prompting | Craft robust briefing prompts for multi-hour autonomous agent runs. |
| memory-systems | Build persistent semantic memory and entity stores across agent sessions. |
| multi-agent-patterns | Coordinate supervisor, worker, and peer multi-agent architectures. |
| project-development | Evaluate whether LLM primitives match product architecture requirements. |
| self-improvement-loops | Build recursive optimization loops that evaluate and refine workflows. |
| tool-design | Design minimal, expressive tool interfaces that reduce agent hallucinations. |
Driving a PR to green, landing a stack, and multi-day autonomous programs are
not available in this repo. They needed Graphite, a real GitHub CLI, and cloud
agents. See parked/README.md for what each would take.