agentleFS
alonbaronalonbaron/claude-skills/skills/architect/SKILL.md

architect

Skill13 starsChanged 34 days agoMITView on GitHub

Produces Alon's design-doc system before code: SOURCE_OF_TRUTH.md, ARCHITECTURE_ROADMAP.md, TODO_WORKFLOW.md, CLAUDE.md, plus a modular docs/architecture set for larger projects. Model first: data and invariants before framework, every invariant enforced at two boundaries, failure paths designed as deliberately as happy paths, one source of truth everything else links back to. Also audits existing docs against the code, drift cited both sides.

174 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 silent 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

9 rules. 0 are common, 6 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 162 edited versions we read keep these.

  • 6

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If a new app, feature, or workstream is starting and no current design docs exist, invoke this without being asked: announce in one line — …

    line 43
  2. Distinctive:

    Keep it tight — it is the contract, not the manual.

    line 57
  3. Distinctive:

    TODO tracks reality: Every task cites the arch §/file it implements.

    line 88
  4. Very specific:

    Definition of "synced": no architectural claim in code that isn't in the docs · no dead cross-links · 00-index File Map matches files on …

    line 90
  5. Very specific:

    Every drift item cites both sides: the file:line in code that states the fact, and the doc (+ § or line) that should describe it and …

    line 100
  6. Distinctive:

    Ask 1–3 blocking questions only (scale, users, hard constraints, existing stack).

    line 146
  7. Distinctive:

    Every component has a defined responsibility *and* a failure mode.

    line 158
  8. Distinctive:

    Verify or say you don't know.

    line 166
  9. Very specific:

    After a phase is implemented → review-swarm the diff before the PR.

    line 180

“Common” is measured: kept by at least 6 in 10 of the 162 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

!ls -1 SOURCE_OF_TRUTH.md ARCHITECTURE_ROADMAP.md TODO_WORKFLOW.md CLAUDE.md docs/architecture 2>/dev/null || true

The line above lists which design docs already exist in this repo, so create-vs-audit mode is known before the first tool call.

Architect

Design and document the system before building it — in the house doc format, kept in sync at all times. Output is documents, not code.

Principles

Enforced in the docs: model first (data + invariants before framework) · enforce every invariant at the core, ideally at two boundaries (app-layer validation and a DB constraint) — never "the frontend handles it" · design failure paths as deliberately as happy paths · small, reversible, independently shippable steps · one source of truth — everything else derives from it and links back.

Proactive use

If a new app, feature, or workstream is starting and no current design docs exist, invoke this without being asked: announce in one line — "Running architect: <why>" — and proceed. Never ask permission to run the skill; the 1–3 blocking questions below are still allowed.

The four artifacts (+ the modular set)

Authority flows top-down. A fact lives in exactly one place and is linked from everywhere else.

  1. SOURCE_OF_TRUTH.md (apex) — the canonical, slow-changing truth: scope & non-goals · the domain model · the invariants and where each is enforced · the load-bearing decisions as mini-ADRs (decision · why · alternative rejected). If anything conflicts with this file, this file wins. Keep it tight — it is the contract, not the manual.
  2. ARCHITECTURE_ROADMAP.md — architecture + phased plan derived from the SoT. Header block (Version · Status · Owner), then §-numbered: 0 Executive context · 1 Tech stack (Layer · Tech · Role table) · 2 Data schema (low-level: columns, types, constraints, indexes, JSONB shapes, decision call-outs) · 3 Backend (structure · services · API-contract table) · 4 Frontend · 5 Execution phases · 6 Non-functional requirements.
  3. docs/architecture/00-index.md + 01-…NN (larger projects only) — agent-friendly modular extracts of the roadmap sections. The index carries a File Map table (# · file · covers · roadmap §), a Dependency Graph (ASCII), and a Quick Reference ("I need to work on X → read these files").
  4. TODO_WORKFLOW.md — the task tracker. Status legend ([ ] · [IN PROGRESS] · [FINISHED - PENDING MERGE] · [MERGED/DONE] · [BLOCKED]); tasks grouped by phase; each row: # · Task · Architecture Ref (linked to the §/file) · Status · Branch. A task closed mid-phase leaves a Left for <id>: ... note for whoever picks it up next.
  5. CLAUDE.md (project root) — the rules file Claude Code auto-loads: operating rules + the sync protocol (template below). One markdown file — no .clauderules, no .cursorrules, no import shim.

Sync protocol — the docs are never allowed to drift

This is the whole point. Bake it into CLAUDE.md and obey it yourself:

  • Top-down, docs-before-code. A change to any architectural fact updates SOURCE_OF_TRUTH.md first (if it touches a truth/invariant/decision), then ARCHITECTURE_ROADMAP.md, then the affected docs/architecture/NN-*.md, then the code. Never ship a change the docs don't yet describe.
  • One fact, one home, many links. A detail is defined once and referenced by link elsewhere. Every doc header links to the others.
  • TODO tracks reality. Every task cites the arch §/file it implements; a task that changes architecture names the doc it updated; statuses are current.
  • Definition of "synced": no architectural claim in code that isn't in the docs · no dead cross-links · 00-index File Map matches files on disk · TODO statuses match git reality.

audit mode

Given architect audit, do not author — verify sync and report drift: code facts missing from the docs, dead links, index/file mismatches, stale TODO statuses. Output a prioritized fix list and offer to apply it.

Every drift item cites both sides: the file:line in code that states the fact, and the doc (+ § or line) that should describe it and doesn't. No item without both is a finding — it's a hunch, and hunches don't go in the list.

CLAUDE.md template (generalize to the project)

# <Project> — Rules

Stack: <one-line stack summary>.

## Commands

| Scope | Install | Test | Lint | Format |
|---|---|---|---|---|
| root | `<cmd>` | `<cmd>` | `<cmd>` | `<cmd>` |
| <package> | `<cmd>` | `<cmd>` | `<cmd>` | `<cmd>` |

Note any package-manager quirk here (workspaces, monorepo tool, pinned version).

## Git (mandatory, no exceptions)
- Open `feature/<topic>` branch BEFORE first edit. Never commit to `main`.
- Micro-commit per logical step. Conventional Commits (feat/fix/refactor/chore/docs/test).
- Commits are authored by the repo owner alone — never add an AI co-author or `Co-Authored-By` trailer, never mention AI in commit messages or PRs.
- Never delete branches. Never force-push. Never skip hooks. PRs only.

## Workflow
1. Locate the task in `TODO_WORKFLOW.md`; mark `[IN PROGRESS]`; state which architecture file you reference.
2. Load `SOURCE_OF_TRUTH.md` + the relevant `docs/architecture/*.md` before coding. Never guess an API surface — verify against version-pinned context.
3. Update status: `[FINISHED - PENDING MERGE]` at PR open, `[MERGED/DONE]` after merge, `[BLOCKED]` with the blocker noted.
4. PR when every task in a phase is `[FINISHED - PENDING MERGE]`.
5. On close, append a dated entry to `docs/handoff.md` (newest-first) — what shipped, what's left, and any `Left for <id>: ...` note for the next task to pick up.

## Reference precedence
- Apex truth: `SOURCE_OF_TRUTH.md`.
- Architecture + phases: `ARCHITECTURE_ROADMAP.md`.
- Modular details: `docs/architecture/00-index.md` (start there).
- Tasks: `TODO_WORKFLOW.md`.
- Handoff log: `docs/handoff.md` (newest entry first).

## Architecture-change rule (sync)
If a task changes any architectural fact, update `SOURCE_OF_TRUTH.md` → `ARCHITECTURE_ROADMAP.md` → the modular `NN-*.md` FIRST, THEN write code.

Before you start

Ask 1–3 blocking questions only (scale, users, hard constraints, existing stack). State assumptions for the rest and proceed — don't stall. Match depth to size: a single feature → SOURCE_OF_TRUTH + a light ARCHITECTURE_ROADMAP + TODO_WORKFLOW + CLAUDE.md; a new product → add the modular docs/architecture/ set. An existing doc set that's just missing one fact — a new invariant, an endpoint, a decision — gets that one addition wired into its right doc and cross-linked; it does not get re-authored from scratch.

Rules

  • No code in this phase. Pseudocode for a tricky algorithm is fine; an implementation is not.
  • Every component has a defined responsibility and a failure mode.
  • Flag unknowns as open questions — never invent a constraint, a number, or an API you haven't confirmed.
  • A choice you made that the user didn't give (a status code, a limit, fail-open vs fail-closed, a library) is a proposal, not a fact: mark it (proposed) in the doc and list it under "Decisions to confirm" in your reply.
  • Decision-dense: tables and bullets, not prose. Call out reversed or forbidden decisions inline (> Do not reintroduce X without an explicit decision).
  • Verify or say you don't know; never invent a path, API, number, or fact. Commits are the repo owner's alone: no AI co-author trailer, no AI mention in messages.

When not to use

  • A small fix or task already covered by current design docs — just do it.
  • Pure implementation of an already-designed phase — code, don't re-spec.

Hand-offs

  • Repo with a remote → run up-to-date first so the design builds on latest code.
  • A load-bearing decision with genuinely competing options → ask-the-council before locking it into SOURCE_OF_TRUTH.md.
  • After a phase is implemented → review-swarm the diff before the PR.
  • The doc set carries prose a human (not an agent) will read — §0 executive context, a README, a mini-ADR rationale → humanizer on those passages only. The tables, invariants, and contracts stay as they are; they're reference text, and "sounding human" is not a goal there.

Done when

The four docs exist, cross-link, and agree with each other and the planned code; a competent dev could build from them without guessing the model, the invariants, the failure handling, or the order — and CLAUDE.md makes the sync protocol non-optional for whoever builds it.

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.