Copy-paste starting points. Nothing here is wired into this repository's own build — these are files you take away and adapt.
| File | What it does | What it does not do |
|---|---|---|
ci/receipt-check.yml |
A GitHub Actions job that fails the build when a delivery carries no gx receipt, or carries one that does not verify offline. It runs gx receipt verify --offline over a receipt, a checkpoint and a trusted-key bundle. |
It is not run by this repository (this repo verifies its own receipts through its test suite instead), and it assumes you supply the key bundle. The file's own header states what it assumes in full. |
demo_one_screen.sh |
A one-screen, 8-stage shell reproduction of commit → receipt → tamper → verify → undo → redo against a real cloned git repository (not a fixture). Run it with bash examples/demo_one_screen.sh. |
It does not exercise the filesystem substrate (git only), and it does not repair the known repeated-undo edge case it discloses in its own header — it just doesn't hit that case, since it only undoes each intent once. |
openclaw-plugin-demo/ |
An OpenClaw before_tool_call plugin (~200 lines) that routes an agent's filesystem writes through the gx membrane: a write can be refused before it lands and taken back after it does, with a third party able to check it happened without the engine, network, or project. Run pwsh -File run-demo.ps1 or node src/demo.ts. |
Not wired into this repository's own build or CI; it is a copy-paste starting point, same as the other rows. |
receipt-memory-export/ |
Derives generic agent-memory nodes (memory/nodes/*, memory/index.*) from gx receipts, using CLI output only (gx plan/gx commit/gx undo/gx receipt show) — never reads .gx/ state or imports a gx-* crate. Run run_demo.sh to drive a real gx project and verify the export against it. |
Verified only against the fs substrate; does not read .gx/ state by design, so it cannot list receipts the way gx replay/gx draft list/gx verdict-checkpoint list each partially can (see its own README for the exact gap). |
README.md |
This page. | — |
Copy ci/receipt-check.yml into your own .github/workflows/, point it at the receipt and key
your delivery produces, and it will refuse a build whose evidence does not check out.
docs/TUTORIAL.md walks the same verification by hand first, which is the
faster way to find out what the three input files are before automating them.
Run demo_one_screen.sh directly (it needs a built gx on PATH, or GX=./target/debug/gx
set beforehand) to watch the same lifecycle end to end against a real npm-package repo.
openclaw-plugin-demo/ and receipt-memory-export/ each carry their own README with their own
run instructions; read those before running either.
This folder is published whole: every file added here later is published too, with no allow-list
filtering it. That is why it holds no staging step to catch a stray file — but it is a folder of
independent starting points, not a fixed count: it held three top-level entries when that sentence
was first written, held four once openclaw-plugin-demo/ landed without this page being updated
to match (caught 2026-09-01, GitDoc inventory pass), and holds five now that
receipt-memory-export/ is listed above. Read the table above for the current count, not this
paragraph's history.
The table above names each top-level starting point and points at its own README for how to run it. This gate also expects every tracked file's basename to appear somewhere on this page, not only the top-level ones, so here is the rest by basename rather than by re-describing each sub-project (their own README does that):
openclaw-plugin-demo/:.gitignore,package.json,package-lock.json,tsconfig.json,openclaw-entry.ts,plugin.ts,membrane.ts,harness.ts,gx-cli-membrane.ts, and underopenclaw-install/:openclaw.plugin.json,package.json, andsrc/index.ts.receipt-memory-export/:export_memory.py.