agentleFS
lohi-ailohi-ai/babysit/.claude/skills/architect/SKILL.md

architect

Skill0 starsChanged 2 days agoMITView on GitHub

Map a system's current architecture and design its next generation with diagrams: verified as-is design, headroom, safety gaps, and a staged scaling roadmap (edge, database, cache, jobs, replicas, services) where each stage waits for a measured trigger. `new` designs the smallest system for a product with no code yet.

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use ecosystem-reference, fractal-delegation and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • naman10parikhSkill

    ecosystem-reference

    Quick-access directory of Claude Code ecosystem repos, skills, plugins, MCPs, and patterns. Use when improving the harness.

    Also in 5 projects that use this

  • naman10parikhSkill

    fractal-delegation

    Also in 5 projects that use this

  • tristanphvnSkill

    ketchup-video

    Catch the user up like /keel:ketchup, with a short subtitled video of slides for what happened and the waiting calls as text under it. Use for /keel:ketchup-video.

    Also in 14 projects that use this

  • jaydubya818Skill

    figure-it-out

    Use when novel or multi-phase work needs an auditable workflow and no narrower playbook fits.

    Also in 55 projects that use this

  • ianclearySkill

    principle-explain-the-number

    Apply before you trust, report, or act on a number you measured: a speedup, a regression, a throughput, a latency, or an eval result. Find what limits it, and rule out that it measured something other than the work you think.

    Also in 29 projects that use this

  • eryzerzSkill

    make-bot-ui

    Use when building a custom UI (page, dashboard, buttons) that should wake a Grok Bot over a webhook, when the user must provide a webhook sender key, or when exposing that UI on Tailscale.

    Also in 20 projects that use this

  • +6 more from naman10parikh

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 177 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Follow preamble and the Auto-Decision Framework.

    line 12
  2. Distinctive:

    Ready to grow means seams, not components: keep the seam now, add the component when its trigger fires.

    line 37
  3. Distinctive:

    Verify: before reporting: every diagram rendered and looked at.

    line 110

“Common” is measured: kept by at least 6 in 10 of the 177 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

architect

System-level design over a product horizon. plan-draft plans one change and team builds a project; this skill decides what the system should become and in what order, and hands each stage to them. It writes documents and diagrams only: never code, infrastructure, configuration or data.

Follow preamble and the Auto-Decision Framework. Shared refs are filesystem paths beside this skill's directory, so read them by path, not as skill://. Detail: roadmap (tracks, triggers, seams) · diagrams (set, homes, render check) · next-gen template.

Modes

  • current — verify the as-is architecture, update its diagrams, audit it. Steps 1–3.
  • next — design the next horizon and roadmap. Runs current first when the current owners were not verified in the last 30 days.
  • No argument — both.
  • new — a product with no code yet: the smallest system that carries the requirement's first release. § New product replaces the flow.

Two levels never mix: current lives in the project's existing architecture owners (project docs), true to code, config and production, nothing speculative; next lives in the next-gen document.

Principles

  1. Simplest that fits: measured load plus the next horizon with headroom, fewest moving parts. Managed and boring wins when operating it costs more than its bill.
  2. Ready to grow means seams, not components: keep the seam now, add the component when its trigger fires.
  3. Every component earns its place: the trigger that justified it, owner, failure mode, monthly cost, how to remove it.
  4. Reversible first: configuration before code before topology.
  5. Attribute before you architect: a load-driven stage needs measured evidence that waste is removed (a perf-audit report or equal).
  6. Split along data ownership: one writer per entity; a boundary cutting across tables two modules write is a distributed monolith.

Inputs and assumptions

  • System: AGENTS.md and the architecture, caching and deployment docs it links; code entrypoints, modules, data access; deploy and infrastructure config; read-only running state. Docs are claims, code and config are evidence. Two docs that disagree: draw what the newest code proves, tag the element (unverified) when code cannot settle it, and open a finding with the command that would.
  • Load: perf-audit reports and the performance profile, else the few numbers the headroom table needs, else unmeasured.
  • Cost: bills or provider pricing pages with the date read; never an invented price.
  • Horizon, first source that has one: PRD → roadmap milestone → open tickets → the owner's stated target → a labelled baseline of ×2 and ×5 current peak.
  • Owner-only facts (growth, budget ceiling, availability target, residency, team size) are never a blocking question: infer, tag assumed (<source>, <date>), and list each under "Your decisions" with its default and the cost if the default is wrong.

Flow

Mirror the steps into the task list.

  1. Inventory what runs: components, stores, caches (each with its owner), queues, scheduled jobs, external services, entry path (DNS, CDN), and where each runs (host, size, region, managed or self-hosted). Mark every roadmap stage of the three tracks in place, partial or absent with evidence.
  2. Current diagrams (diagrams), each in its home. In current owners correct only facts verified wrong and the diagrams; trimming, moving or other drift is a finding with its target owner, as in project docs § Audit.
  3. Audit. Findings carry evidence, severity (critical outage or data-loss risk now · high breaks at the next horizon · medium taxes every change · low) and the smallest fix:
    • Safety floor gaps (always critical).
    • Headroom per component: peak, limit, headroom ×, what saturates first.
    • Single points of failure and what fails together, against the availability target.
    • Coupling: tables written by several modules, synchronous chains, cycles, shared mutable state.
    • Scaling cliffs: in-process state, local disk, singleton jobs without a lock, unbounded tables or queries, fan-out per request.
    • Cost: spend per component against what it serves.
    • Operability: rows of agent-operable the system does not meet (UI-only operator actions, writes with no preview or retry safety, no health or needs-attention read path), high when an agent is due to run the product within the horizon.
  4. Horizon: name it and its drivers, each new feature's load shape (read-heavy, write-heavy, CPU-bound jobs, real time, large files) and multiplier; project the headroom table and order the limits crossed.
  5. Next-gen design: the minimal change set that clears those limits and carries the new features. Diagram it with changes marked, say what stays the same, record each costly-to-reverse decision.
  6. Roadmap: the stages the horizon or a finding needs, in roadmap order, each with a trigger in the <signal> <op> <threshold> in <n> of <m> windows form, prerequisites, change, what to watch, cost, rollback and how to verify; the seams to keep now; a "not yet" list. A trigger whose signal is not collected adds it to the profile as TODO(2build):.
  7. Verify before reporting: every diagram rendered and looked at; every current fact has a file, command or console source; numbers carry date and source; facts that could not be read are in "Verify next".

New product

Nothing runs and nothing is measured, so steps 1–6, headroom and the next-gen document do not apply; principles 1–4 and 6 do. Input is the requirement (docs/REQUIREMENT.md or the caller's) and its first-release subset.

  1. Shape: the fewest deployable services the first release needs. The starters split a product into a web app that renders and an API that owns data: a first release that keeps data on a server takes both, one that keeps none takes the web app alone. Never build a data layer inside the web starter. Each service takes its stack from a starter template when one fits; a part no template covers (native mobile, desktop, a CLI) is named as outside the starters with the generator it will use.
  2. Data: the entities, the one writer of each, the one store. Prefer what runs on the owner's machine without an account.
  3. Where it runs: this machine now; one production target named with a monthly price from the provider's pricing page and the date read; a price that cannot be read is a TODO(2build): line and "not known yet" in the owner summary, never an estimate. Nothing is created, bought or deployed.
  4. Write the system document at docs/architecture.md from project docs's headings, with one rendered containers diagram. Key decisions also list what was deliberately left out, at most five, each with the signal that would add it. A fact decided later is a TODO(2build): line.
  5. Owner summary, returned in the reply to owner language and kept out of the document: what the product is made of (roles, not component names), what it will cost per month once live, what was left out and when it would be added, and each decision that is the owner's with its default.

Output

The owner summary comes first, in the owner's language (next-gen template § For the owner; § New product step 5 in new mode).

STATUS: DONE | DONE_WITH_CONCERNS | NEEDS_CONTEXT | BLOCKED
VERDICT: ARCHITECTED(<current|next|both|new>, findings=<n>, stages=<n>)
DOCS: <updated current owners; next-gen doc; rendered diagram paths>
SUMMARY: <biggest risk, first limit to break and when, first stage and its trigger>
NEXT: <owner decisions; /bbs:autopilot "<seam or critical fix>"; /bbs:team "<stage>" for a phased stage; /bbs:perf-audit when load is unmeasured>

An unrendered diagram is a concern, not a deliverable. A next-gen design without one measured or sourced load number is DONE_WITH_CONCERNS, every projection tagged as an assumption.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.