c0py/docs/AUTOMATIONS.md

C0PY Automations

Automation principles

C0PY automations are resumable, idempotent, versioned and evidence-producing. Every run emits a manifest, provenance and machine-readable failures.

Canonical pipeline

DISCOVER
  -> CAPTURE
  -> NORMALIZE TO IR
  -> MODEL PRODUCT
  -> FREEZE SPEC
  -> RECONSTRUCT
  -> DEPLOY PREVIEW
  -> VERIFY
  -> REPAIR (if failed)
  -> VERIFY
  -> RELEASE CANDIDATE

Jobs

discover-target

Inputs: target URL, capture profile, scope policy. Outputs: route graph, navigation graph, discovered workflows, crawl queue.

capture-target

Uses Playwright/CDP and optional adapters. Captures DOM/runtime HTML, styles, accessibility, media/assets, screenshots, network/HAR, storage metadata, events and performance.

interaction-sweep

Enumerates deterministic interactive states and records before/after deltas. Semantic agents may propose candidate flows; deterministic capture verifies them.

workflow-discovery

Uses semantic browser adapters to discover product tasks and translates them into explicit action/state graphs with evidence links.

normalize-evidence

Validates captured artifacts and writes the Canonical IR/Evidence Graph. Rejects untyped or provenance-free critical evidence.

generate-spec

Creates frozen reconstruction contracts for routes, components, design tokens, responsive states, workflows and network/domain behavior.

reconstruct

Dispatches versioned specs to selected builders (v0, lovable, codex, generic). Generated code is never treated as evidence about the source.

preview

Builds and deploys a preview target. Vercel is a supported deployment adapter; SIAX/Coolify remains supported for controlled infrastructure.

verify

Runs the fidelity vector: visual geometry/style, responsive, routes, interactions, workflows, network, motion, accessibility, runtime and build checks.

repair

Turns verification failures into typed repair tasks, applies the smallest justified patch and sends the result back through verification. Repair may not relax gates without an explicit versioned policy change.

upstream-reconcile

Scheduled/manual automation that checks registered upstream projects for high-value changes. It creates candidate capability records only; it never auto-merges foreign code without provenance/license/test gates.

schema-compatibility

On every IR/schema change, validates backward compatibility and requires a version bump for incompatible changes.

license-provenance

Checks all CHERRY_PICK and vendored/adapted source entries for repository, SHA, source path, classification and license/notice requirements.

Required run manifest

Each automation run records at minimum:

• run id
• target id
• C0PY commit/version
• schema version
• capture profile
• started/completed timestamps
• tool/adapter versions
• evidence object ids
• warnings and typed failures
• retry lineage
• output spec version
• verification result

Failure policy

No automation may convert a failed measurement into an inferred success. Failed or inaccessible states remain explicit evidence gaps.

CI/CD gates

Pull requests touching capture, IR, product model, reconstruction contracts or fidelity rules must run:

1. typecheck
2. unit tests
3. schema validation
4. provenance validation
5. license boundary validation
6. representative deterministic fixture capture
7. reconstruction fixture verification when affected

Production promotion is blocked when critical route/workflow/state gates fail.