Architecture · LikeC4

coding-agent-workflows

Renders one neutral source layer into a native practices bundle per coding agent, and installs it into consuming projects

built evolving planned research risk

The whole picture

System landscape

The system in context, then opened up into its containers.

coding-agent-workflows — system landscape

coding-agent-workflows — system landscape

index
coding-agent-workflows — containers (source → build → targets → install)

coding-agent-workflows — containers (source → build → targets → install)

cawSystem

Renders one neutral source layer into a native practices bundle per coding agent, and installs it into consuming projects

Risks & open questions

Risks & open questions

risks
Planned & in-flight work

Planned & in-flight work

planned

Inside each box

Containers & components

Each part decomposed into the components that implement it. Every box links to its source in the interactive explorer.

source/ — authored bundle (the only hand-edited layer)

source/ — authored bundle (the only hand-edited layer)

sourceContainer

The single source of truth: rules, agent roles, skills, workflows, templates, and the scope manifest. Everything else is generated from here

build/ — renderer + staleness & release gates

build/ — renderer + staleness & release gates

buildContainer

Mechanism only (IO + template substitution, no content decisions): turns source/ into targets/ and AGENTS files, then gates the rendered output

targets/ + AGENTS — rendered per-agent output

targets/ + AGENTS — rendered per-agent output

targetsContainer

Pre-rendered and committed; no build step before use. The artifacts install.sh copies into a consuming project

scripts/fleet/ — machine-level conformance track

scripts/fleet/ — machine-level conformance track

fleetContainer

A separate machine-level track (install.sh fleet): passively register every repo and scan it for guardrails/testing/observability conformance

How it runs

Walkthrough flows

Dynamic views — the narrative spine of the system, step by step.

Walkthrough — render source/ into every target

Walkthrough — render source/ into every target

buildFlow

Mechanism only: render reads the scope manifest, resets targets/, and emits the thin AGENTS.md index, the full AGENTS.full.md bundle, and the native Claude + Codex layouts — injecting the generated rule catalog into the Claude coding-practices skill so it can never drift.

Walkthrough — staleness + release gates

Walkthrough — staleness + release gates

releaseFlow

The repo`s one mechanical invariant: committed output must match a fresh render of source/, and must carry no absolute paths, PII, or internal jargon. CI and the pre-commit hook both run these; a failed staleness check restores the committed snapshot so it never leaves rendered changes behind.

Walkthrough — install into a project and consume it

Walkthrough — install into a project and consume it

installFlow

A plain file copy — no hooks, no daemons. The consumer picks its agent`s layout; the agent then auto-loads the thin index and reads only the section a task needs, keeping always-loaded context small.

Where it runs

Deployment

What runs where, and the process & data boundaries between the pieces.

Deployment — where each piece runs

Deployment — where each piece runs

deployment

Explore it live

The figures above are static exports. The interactive explorer lets you pan, zoom, follow relationships, and jump from any box to the source.