agentleFS
kaotyprkaotypr/context-circuit/.agents/skills/cc-system-design/SKILL.md

cc-system-design

Skill0 starsChanged 17 days ago0BSDView on GitHub

Author a system design as source material under sources/system-design/ — the three-tier layout, how much detail belongs in each file, and why scopes separate by concern rather than by repository. Drafts structured source files only; never approves, plans, executes, or writes knowledge.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    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:

    Use this on a request to work out the shape of something larger than a single change — "design the system for X", "write up the …

    line 8
  2. Very specific:

    Every folder carries a README.md index.

    line 53
  3. Very specific:

    Every fact lives once. If a reviewer must open five files to grasp the design, too much left design.md.

    line 83
  4. Very specific:

    Start minimal, with a README.md and a design.md, and grow from there.

    line 94
  5. Distinctive:

    Draft structured source files only: Never approve, accept, plan, execute, merge, deliver, or write knowledge.

    line 138

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

Design a system before building it

Use this on a request to work out the shape of something larger than a single change — "design the system for X", "write up the architecture for this feature", "structure the design before we build it". It is for the developer who wants the whole picture settled, and legible, before any of it is implemented.

What it produces is source material: passive, durable, written to be read by a person deciding whether the shape is right. It has no status, no acceptance gate, and no runtime record. Drafting a design is not approving one.

A bounded change does not need this. Go from the project's knowledge straight to an intent and a plan; a design with five near-empty files is worse than no design at all.

Where it goes

Author under sources/, which the workspace treats as passive evidence:

sources/system-design/<product-or-project>/<grouping>/<scope>/
  • <product-or-project> names the product as a whole — never a single repository. In a single-product workspace this level may collapse to sources/system-design/<grouping>/<scope>/.
  • <grouping> is the organizing dimension for a body of design work. Default: a version — 3-number semver with a v prefix (v2.0.0, not v2.0). A named grouping — phase-2, Q1, a milestone slug — is a deliberate choice for a team that does not organize design by release. Within one product, pick one dimension rather than mixing release and phase folders.
  • <scope> is a concern.

Where the author does not name a grouping, derive it from the version the design targets: the project's next version for forthcoming design, which is the normal case, or its current version when documenting as-built state. Read that version from the project's own source of truth — whatever its repositories actually use to version themselves. Never take it from workspace.yaml, which records the Context Circuit schema and not this project's releases, and never invent one. Where the targeted version is genuinely ambiguous, surface the choice rather than guessing.

Writing these files is an ordinary authoring write. Afterwards sources/ is passive again: it is read only when a request or a record names a file in it.

The three-tier layout

Every folder carries a README.md index; each scope's normative content is design.md; detail splits into files or sub-folders as it grows.

sources/system-design/<product-or-project>/
  README.md                 # index of groupings (landing + reading order)
  <grouping>/
    README.md               # grouping index: the scopes and the reading order
    <scope>/
      README.md             # scope index
      design.md             # overview — NORMATIVE, readable end to end
      <concern>.md          # one concern each
      <concern>/            # a concern with parts: README + a file per part

How much detail per file

Calibrate by what a reader needs from the file, not by filling it:

  • README.md — orientation only. What exists here and in what order to read it. No design content, decisions, or mechanism.
  • design.md — the normative overview a reviewer can stop at. The capability, the problem, the principles, the fixed decisions, and the shape of the whole — readable start to finish without opening a single detail file. It states what and why, and defers how to the detail files it links.
  • <concern>.md — one concern at implementation depth. Its mechanism, interfaces, edge cases, and the diagram that explains it. Name it for the concern (scheduling.md), never details.md. It continues from design.md and never restates it.

Every fact lives once. If a reviewer must open five files to grasp the design, too much left design.md; if design.md runs many screens of mechanism, too much stayed.

Split only when a concern earns it

  • A one-paragraph concern is a paragraph in design.md, not a file.
  • A concern that outgrows a section becomes its own <concern>.md.
  • A concern with distinct parts becomes a sub-folder — a README.md index and one file per part.

Start minimal, with a README.md and a design.md, and grow from there. Do not pre-fragment.

Separate scopes by concern, never by repository

A system design earns its keep by describing how the pieces fit across the system, so scopes cut by concern — a backend topology, a domain, a client application, a cross-cutting flow — matching the product rather than the git layout.

  • Never one scope folder per repository. That fragments exactly the cross-repository coherence the design exists to capture.
  • Where one repository needs deep design detail, that is a scope within the product's design — a backend/ scope, say — not a separate top-level design and not a new <product-or-project>.

Diagrams

Embed diagrams as fenced ```mermaid blocks in the file they explain. That is the durable form: it renders on GitHub and in most editors, and it survives in the file rather than in a tool. You may preview through a host plugin where one exists, but never depend on one — with no plugin, still emit the fenced mermaid.

How it reaches the rest of the workspace

A design is source material, so it feeds the workspace the way all source material does, with nothing new added to the flow.

Knowledge. What the design settles durably becomes a note in context/, written through .agents/skills/cc-knowledge/SKILL.md. The note carries the concept in its own words and anchors to code; it never cites the design file, because sources/ is archived while knowledge outlasts it. This skill never writes knowledge itself.

Intents. An intent may name a design file when the outcome it asks for actually depends on one — a path under sources/system-design/, cited where it bears on the goal, the constraints, or the repository scope. Only where it is needed: most intents settle their outcome in their own words and name nothing. An intent is still written before detailed code investigation, and a cited design does not replace reading the real code after approval. There is no second approval of the design, and citing it does not make it a gate.

Boundaries

  • Draft structured source files only. Never approve, accept, plan, execute, merge, deliver, or write knowledge. This skill grants no route, role, or authority.
  • No status, no gate, no runtime record. A system design is a source.
  • One home. Design material lives under sources/system-design/ and nowhere else. Do not write design files under an intent's directory.

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.