Proofline 0.2.0-rc.1 — evidence for the current task

Proofline 0.2.0-rc.1 — evidence for the current task

A passing check can belong to an earlier version of the task. This preview makes that context queryable: see the current gaps, recheck the affected evidence, and retain observations whose declared dependencies still match.

What changed

• Machine queries: complete, read-only goal-delta --json and compact query, with gap/status/item/evidence filters and explicit observation/target revisions.
• Inspectable configuration: doctor identifies missing input scope, broken mappings and uncovered requirements; init suggests input paths that actually exist.
• Ordinary evidence with context: run and capture bind observations to the selected contract and dependencies. Unchanged dependencies can retain their earlier receipts and observation revisions.
• A shorter Intake handoff: a reviewed intake-requirements/1 file can be passed directly as the contract. The separate import-intake command remains available when a materialized contract is useful.
• Usable recheck proposals: Goal Delta returns executable, argv and cwd, carrying the selected manifest, target, dependencies and target-digest precondition. The returned command and file-capture actions were actually executed in package checks. A changed target is rejected before observation.

proofline doctor --contract reviewed.json --dependencies dependencies.json --json
proofline run AC-01/tests --contract reviewed.json --dependencies dependencies.json -- node --test
proofline query --contract reviewed.json --dependencies dependencies.json --gaps
proofline goal-delta --from prior-reviewed.json --to reviewed.json --dependencies dependencies.json --json --gaps

Configure the project's manifest and explicit dependency map first; see machine queries and the Intake handoff. For programmatic rechecks, use the argument array with shell: false; the legacy action.command field is now null. The tool does not execute proposals automatically.

Install and compatibility

Download the attached TGZ for an installable Node.js 20+ CLI, or the source ZIP for a checkout-style installation. Verify it with its SHA-256 sidecar. The package, CLI and plugin manifest identify 0.2.0-rc.1.

Intake 0.3.0-rc.1 supplies atomic explicit review batches and saved requirements. Workprint's existing 0.3.0-rc.3 release accepts the public projection; its runtime does not need an upgrade for this flow.

Verified scope

Source: 033903f6602b6f490e93df6d9364ec56b5ae35f3.

• Exact-source CI: all 9 Windows/macOS/Ubuntu × Node 20/22/24 jobs passed, including actual npm packing and installed-package CLI checks.
• Local developer checks: 55 tests, lint, the 42-file release gate, 10 Windows implementation/plugin-identity checks, source/package CLI loops, and 1280/390px report checks.
• Independent owner checks used the final Proofline TGZ, extracted Intake ZIP and the already-public Workprint TGZ. Atomic error handling, direct snapshots, non-default manifest/different caller directory, returned run/capture actions, target-drift refusal, unchanged-receipt reuse and public projection build/verify passed.
• The diagram is an editorial rendering of that synthetic installed-CLI example, not a real-user trial. Its editable SVG and compact provenance accompany the PNG.

The five states retain their meaning. An observed exit code does not establish business-test adequacy; a receipt does not authenticate its author. Local query/action details can contain selected paths and arguments. Workprint's separate public projection excludes those fields and Intake acceptance text.

The retained internal replay was 14 baseline / 29 assisted operations. This revision reduces the assisted sequence to 22, with the same baseline and correctness criteria. The full synthetic CLI loop drops from 17 calls to 14. These results do not establish human-time, agent-time or token savings; see the complete comparison.

This remains a prerelease. Full physical-Mac/fresh-task plugin activation and external-user benefit are unverified for this version.

Proofline v0.1.0 — Proof, not promises.

Proof, not promises.

Proofline turns acceptance criteria into an inspectable graph of observed commands, files, screenshots, environments and timestamps. This is the first public preview.

What's included

• Five explicit evidence states: verified, missing, stale, declared-only, failed.
• Gap Tour: walk each unsupported proof line, inspect its stored reason, and navigate with G / Escape. This release fixes the inspector remaining visible after Exit.
• Goal Delta: compare two contract revisions, follow explicit dependencies, and see which green receipts become stale while unrelated evidence stays usable.
• Self-contained HTML, Markdown and JSON reports; deterministic workprint-profile/0.1 Goal Delta export.
• A dependency-free Node.js 20+ CLI, optional Codex Skill/Stop hook, and bundled synthetic examples.

Try it

Download codex-proofline-0.1.0.tgz and run:

npm install --global ./codex-proofline-0.1.0.tgz
proofline version
proofline init

Or extract codex-proofline-0.1.0-source.zip, open an example HTML report, or run node bin/proofline.js. Use the source ZIP for development and the standalone Mac gate. Each archive has a SHA-256 sidecar. Distribution is through GitHub Releases; no npm registry publication is implied.

Observed release checks

• Source commit: 45eaebd580026a3fc2a56c1e072bf50234041d67.
• Six hosted CI jobs passed: Windows/macOS × Node 20/22/24, including actual npm pack --dry-run, installation and packaged CLI/example smoke.
• Windows: 42/42 core tests; lint; 42-artifact release gate; official plugin/Skill validators; isolated implementation check. A clean Git clone also passed the core tests and release checks.
• Browser: 13 passing interaction checks, desktop and 390×844 views without overflow, using synthetic examples.
• The final TGZ was packed twice with identical bytes and installed locally; the CLI loop, five states, three report formats and all three exact Goal Delta projections passed.

Receipts are local and tamper-evident, not attestations. Hosted macOS CI does not establish physical-device or fresh-task Codex integration evidence. The stronger standalone real-Mac gate remains pending for this candidate; the historical targeted Lab PASS belongs to the prior exact candidate. Real-user and production validation are unobserved. Full evidence record.