Two local-first tools for field engineering, pinned together and verified as one suite.
Field engineering runs on two artifacts: the deck you present and the evidence that says the pilot worked. I built one tool for each and kept them separate on purpose. A customer can take deckforge without proofpack. An agent can drive any of them alone. fieldpack is where the seams get tested. It pins the two as submodules and runs a verification suite across them. Does a proofpack export import cleanly into deckforge and fit on every slide? Do the browser UIs pass an axe accessibility audit at desktop and phone width? Does each product still install from its packed tarball without its siblings?
| Tool | Does |
|---|---|
| deckforge | Presentation studio with a measured headless preflight |
| proofpack | Pilot criteria, evidence and customer-safe handovers |
Both share one shape: a CLI and an MCP server over a JSON workspace you initialise explicitly, revision-checked mutations, --read-only mode, and no model calls, uploads or telemetry.
Node 23 or newer.
git clone --recurse-submodules https://github.com/ong6/fieldpack.git
cd fieldpack
npm run doctor # each submodule matches its pin, dependencies present
npm run setup # npm ci in every product and in verification/
npm run browser:install # Chromium for the headless and browser tests
npm run test:products # the two product suites
npm test # doctor, then the 23-test cross-product verification suite- Agent pipeline: proofpack and deckforge driven end to end through their product APIs in one temp workspace.
- Contract: proofpack's deckforge export derives only from customer-visible records and never changes when private fields do.
- Exports: an offline HTML deck from a proofpack readout, measured for overflow.
- Browser: edit, persist, reorder, import, export, present and print in each UI, with axe audits and screenshots at desktop and mobile width.
- Packaging: every product installs from
npm packoutput with no sibling present.
Forges make things, packs bundle them.
- groundplane — fails the build when an agent asserts a fact its tools never produced
- jobforge — grades the interview plan you say out loud, not the code you submit
- skillsmith — makes an agent skill from your repo, then proves it beats no skill
- deckforge — agent-first presentation studio with a measured preflight
- proofpack — pilot evidence, review proposals and customer-safe handovers
- skillpack — the Claude Code and Codex skills used across all of these
- uipack — React and SVG figure components behind the diagrams on junxiong.dev