AGENTS.md · git:20261030.eb2f833 · 2026-10-30 · sha256 f9b223bc9c529d04
AGENTS.md git:20261030.eb2f833A
Immutable. This exact content never changes and is served at /api/v1/blob/f9b223bc9c529d04.
# AGENTS.md Instructions for coding agents working in this repository. People should read [CONTRIBUTING.md](CONTRIBUTING.md); this file is the short version of it. ## Layout - `skills/measured-type/`: the skill (SKILL.md, `references/`) and its tools (`scripts/`, tests in `scripts/test/`). Run commands from here. - `specs/`: one numbered spec per change, with requirement IDs. - `evals/`: eval cases, graders, analysis scripts and committed results. ## Commands ```bash cd skills/measured-type npm install && npx playwright install chromium # once npm test # the whole suite node --test scripts/test/<name>.test.js # one file ``` From the repository root, `npm run gate` runs the prose check and every test; it is what "done" is checked against. Set `MT_PYTHON` to a Python that has fontTools, or four tests skip. ## Rules - **Test first.** Add the requirement to a spec in `specs/`, write the test named with its ID, watch it fail, then change the code. Run the whole suite before you say a change works. - **Never approve what was not computed.** A check that cannot run reports under Not verified; it never passes in silence. - **Do not edit anything under `evals/results/`** except to add a new experiment. Do not change a grader after the runs it grades. - **Do not touch `.github/brand/`** or the license files. - **Claims need evidence.** Every rule, number and statement in SKILL.md and `references/` is either measured by a script or cited (Bringhurst section, WCAG criterion, specification URL). Paraphrase sources; do not paste them. - **Style for prose:** no em dashes, no middle dots, no comma before the "and" or "or" that ends a list, British spelling outside code. ## Done means `npm run gate` passes from the repository root with nothing skipped that was not skipped before, and the change, its spec and its tests are in one commit.