Docs as memory

copperhead has no hidden database. Its memory is the markdown in your repo, which means you can read all of it, edit all of it, and review it in a pull request like anything else.

What init scaffolds

copperhead init reads your schematic and writes these into the docs directory (docs/ by default):

File	What it holds
SPEC.md	What the device is, plus top-level constraints and budgets
BOM.md	Bill of materials
PINOUT.md	Pin assignments
SUBSYSTEMS.md	The design broken into functional blocks
LAYOUT.md	Layout intent: what has to be near what, and why
DECISIONS.md	Append-only decision log
CHANGELOG.md	Per-run narrative of how the design got here

DECISIONS.md and CHANGELOG.md are append-only. The agent adds to them and never rewrites history.

Budgets are constraints

Hard numbers in SPEC.md are not decoration. Budgets declared in .copperhead/config.json are surfaced verbatim into every run’s system prompt, and a change that would blow one is refused rather than quietly accepted.

{
  "budgets": { "sleep_current_uA": 25 }
}

Ask for something that costs more than the budget allows and the agent says no and explains the arithmetic, instead of shipping the change and letting you find out at bring-up.

Drift is a build failure

Docs that disagree with the schematic are the failure mode this whole tool exists to prevent. copperhead check compares the two and reports every mismatch in the form “this doc claims X but the actual value is Y”. Because check runs with no LLM and no network, you can put it in CI and in a pre-commit hook, and the hook is installed for you by init.

Hand edits that desync the docs, the constraint registry, or the schematic fail at commit time. See Verify and sync.

.copperhead/README.md

The config directory describes itself. init writes a README into .copperhead/ documenting every key in config.json and what constraints.json is for, so someone who clones the repo and has never used copperhead can still tell what these files are.