Process completed with exit code 1.

AssertionError: README has no paragraph mentioning renaming a spec file: expected undefined to be truthy - Expected: true + Received: undefined ❯ tests/file-scoped-ids.test.ts:1339:88

AssertionError: expected '# 2119\n\n**Spec-driven test enforcem…' to match /### N: Title[\s\S]{0,400}canonical[\…/i - Expected: /### N: Title[\s\S]{0,400}canonical[\s\S]{0,100}<stem>\.N\.M|<stem>\.N\.M[\s\S]{0,100}canonical[\s\S]{0,400}### N: Title/i + Received: "# 2119 **Spec-driven test enforcement for coding agents.** Named for [RFC 2119](https://www.rfc-editor.org/rfc/rfc2119) · [unsupervised.com/2119](https://unsupervised.com/2119) 2119 makes the planning → building → testing loop hard to cheat: 1. **Plans become requirements.** Features start as specs in `specs/` — RFC 2119 documents where every requirement is a numbered, individually addressable statement with exactly one normative keyword. `2119 lint` enforces the format. 2. **Requirements become tests.** Every MUST-level requirement needs at least one test annotated with its ID (`// 2119: REQ-001.2.3` — a comment, so it works in any language). `2119 cover` fails on any gap, in either direction. 3. **Tests get judged.** `2119 review` generates one instruction file per requirement asking a *fresh-context* reviewer — one that did not write the code — a single question: **would these tests fail if this requirement were violated?** Verdicts are recorded with `2119 pass` / `2119 fail`. 4. **One gate.** `2119 check` = lint + coverage + verdict freshness (a verdict counts only while its hash still matches the content it vouched for). Exit code 0 or it isn't done. Agent hooks, git, and CI all call the same command. Here is the gate catching an untested requirement and a stale review: ```console $ npx rfc2119 check specs/session-handling.md:12 [REQ-002.2.4] session-handling.1.2 (MUST NOT) has no covering test annotation: \"Expired sessions MUST NOT be resumable with a stale token.\" .2119/verdicts/session-handling.1.1--55e7c65a2f4f.json:1 [REQ-003.3.1] session-handling.1.1 has no current review verdict (review ID session-handling.1.1--55e7c65a2f4f); run `2119 review` check: FAIL — 2 violation(s), 1 uncovered, 0 failing review(s), 1 stale review(s) $ echo $? 1 ``` This repo practices what it enforces: 2119's own requirements live in [`specs/`](specs/), every MUST has an annotated test, and [`.2119/verdicts/`](.2119/verdicts/) holds the committed review verdicts. To see the full gate running in another codebase — specs, committed verdicts, and the CI check on every PR — browse [panopticon](https://github.com/tylerwillis/panopticon). Three things 2119 is deliberately **not**: a test runner (`check` never executes your suite — compose them: `npm test && npx rfc2119 check`), a CI replacement (it's one exit code your CI calls), and a security boundary (a deliberate cheater is made *conspicuous*, not impossible — see [Risks](#risks)). These boundaries are [enforced as reviewed requirements](specs/REQ-008-honest-boundaries.md); the reasoning lives in [docs/design.md](docs/design.md), and [docs/scaling.md](docs/scaling.md) covers hardening for larger or more formal projects. ## Use it in your repo From your project root: ```bash npx rfc2119 init # the core: specs/, .2119.yml, AGENTS.md section npx rfc2119 init --agent claude # + hooks and a reviewer subagent (also: codex, gemini) npx rfc2119 init --git-hook --ci # + pre-commit gate and GitHub Actions backstop ``` (`npx` works in any repo with Node available, whatever language the project is. Node projects can pin the version with `npm install -D rfc2119`.) Then, to work: 1. Have your agent plan each feature **as a spec** in `specs/` — `2119 lint` keeps the format honest. 2. Build. Every MUST-level requirement needs a test annotated with its ID (`// 2119: REQ-001.2.3`). 3. Run `npx rfc2119 check` (alongside your test runner — `npm test && npx rfc2119 check`, both, always). Fix lint/coverage failures directly; for pending judgment reviews, run `npx rfc2119 review --dispatch` — it emits a prompt you paste into your agent session, and the agent sends each instruction file to a fresh-context subagent (a clean session that di