Proofloop 2.1.1

Proofloop 2.1.1

• Valid FAIL reviews and open findings now report REVIEW_REJECTED and remain rejected on later status calls. Malformed, mismatched or stale responses report REVIEW_INVALID.
• finish refreshes status.json, including on rejection. New execution attempts remove old status/finish snapshots.
• Clarify the checker return task ID and remove the redundant status round trip after finish. No new reporting requirements.

Update the bundled skill:

npx skills add regenrek/codex-proofloop --skill proofloop -g -a codex -y

Standalone CLI:

npm install -g codex-proofloop@2.1.1

Finish active runs with their installed version before updating; changed runner code invalidates old execution evidence.

Verified through real Git/subprocess CLI acceptance, package installation/version lookup, TypeScript build, Oxlint, Oxfmt and skill validation.

Proofloop 2.1.0

Proofloop 2.1.0

Fixes from the first real project pilot, with less manual reporting.

• Hash direct local check scripts and declared helper/fixture inputs, even when ignored. Changed or removed inputs invalidate evidence and review.
• Run Vitest JSON and supporting build/lint/typecheck commands directly, without Node-test wrappers. Process checks claim zero tests and cannot replace behavioral checks.
• Preserve counts for negative TAP reports; distinguish failed tests, malformed reports, empty artifacts and missing artifacts.
• Add --version and opt-in --compact output linking complete records on disk.
• Use compact checker handoffs, one focused review and targeted fix reviews. No mandatory success report, feedback diary or duplicate full-suite review.

Install or update

npx skills add regenrek/codex-proofloop --skill proofloop -g -a codex -y

The skill includes the CLI. For standalone CLI use:

npm install -g codex-proofloop@2.1.0
proofloop --version

Finish active runs before upgrading. Existing 2.0 policies remain supported; execution evidence needs rerunning under 2.1 because the runner and input binding changed. Ignored transitive helpers must be declared in inputs; no automatic import-graph discovery.

Verification

Public CLI acceptance in real temporary Git repositories; Node 24/26 CI. Actual Vitest 5.0.1 pass/fail/skip/todo/zero-test runs, copied-skill version lookup, isolated tarball installation, TypeScript build, Oxlint, Oxfmt and skill validation. No runtime npm dependencies added.

Proofloop 2.0.0

Proofloop 2.0.0

A smaller native Codex workflow with a TypeScript CLI, actual execution evidence, and GPT-6 Luna/max as the default independent checker.

Install

npm install -g codex-proofloop@2.0.0
proofloop --help

Without a global installation:

npx --yes codex-proofloop@2.0.0 --help

Install the Codex skill:

npx skills add regenrek/codex-proofloop --skill proofloop -g -a codex -y

Requires Node 24+ and Git. The CLI has no runtime npm dependencies.

Changes

• Four Python variants replaced by one native Codex skill and a TypeScript CLI.
• Herdr, Sentinel, Fable and Planr coupling removed.
• One frozen project policy replaces duplicated profiles, phase contracts and test-line budgets.
• Real process results, Node TAP / Playwright JSON reports, fresh artifacts and Git working-state fingerprints determine completion.
• Intermediate commits and staged changes remain visible; failed, skipped, zero-test and stale checks cannot finish.
• Explicitly justified test retirement is supported.
• Native GPT-6 Luna/max review is the skill default; reviewer attestations are tied to the checked evidence and are explicitly not authenticated by the local runner.
• Oxlint and Oxfmt are enforced in CI, including consistent formatting of bundled JavaScript.

Breaking changes

The old skill names and Python entrypoints are removed. Finish existing runs with the previously installed version, then install proofloop and create a new proofloop.json. Existing v1 profiles/contracts/run records are not migrated.

The CLI executes configured checks. Native task coordination requires the Codex skill, available host tools and applicable user authorization. This remains an experimental execution guardrail, not a filesystem sandbox or a claim of complete product coverage.

Verification

• 22 real Git/CLI acceptance scenarios.
• Lint, formatting and TypeScript build.
• GitHub CI on Node 24 and 26.
• Isolated installation and CLI startup from the exact release tarball.

Codex Herdr Orchestrator v1.0.0

1.0.0 — 2026-08-11

First public release.

• Three standalone skill variants: Sol + Luna, Sol + Luna + Fable, and Sol + Luna + Fable + Planr.
• One-writer workflow with optional read-only review and monitoring.
• Opt-in, time-bounded silent sentinel with durable run evidence.
• Deterministic Git-based test gate that keeps exploratory tests temporary by default.
• Focused validation, permanent-test admission, and full-suite safeguards.
• Dependency-free runtime scripts and project templates in every variant.
• Automated tests and GitHub Actions for Python 3.11 and 3.13.