Frontier-simplify 2.3.0

Frontier-simplify 2.3.0 checks every finder review with two independent validators and lets the calling agent finish every loop without waiting for a person.

• Finder plus two validators (#56, #59). When a review would be recorded, a refuter and a reproducer run in parallel over the finder's whole answer. Each gets its own fresh checkout with the same inputs and does not see the other's answer, and no further model combines them. They are skipped only when a first review's last nonblank line is exactly Validation candidates: NONE. A failed, empty or timed-out validator returns 5 with an intact receipt, and the finder's review still feeds the next follow-up through PRIOR_VALIDATION.md and copies of the earlier answers.
• Cancellation ends as cancelled. SIGINT and SIGTERM stop every validator, keep what each one already wrote and release the lock. The attempt uses one reservation, and a cached auto does not rerun it.
• No redesign stop (#59). The runner no longer returns 12 or writes REDESIGN-REQUIRED.md. A key that stays open in two consecutive reviews only adds a note, and the review runs anyway. Failing witnesses (13) and the three-attempt limit are unchanged, and old records stay readable.
• The calling agent finishes (#51, #53). SKILL.md, HANDOFF.md, status and the READMEs give a next action for each exit. Every path ends in the existing completion path or as failed with its cause. A clean 10 on the exact head continues to completion, and a review whose inputs changed before it finished is never offered as complete.
• Prompt adjustments (#56). A style point alone never forces a repair. Each finding says whether this change introduced it. Reported or resolved findings are never re-issued under a new ID.

None of this adds a judge, a vote, a result schema, a new ledger version or a retry policy, and no exit status or validator agreement authorizes a merge.

Verification

Release commit: 1dcc72ed7d16d24d1f338f169f585094f32a6c56 (merge of #68; its tree is identical to the PR head 3cddb66).

• PR checks ran the full offline selftest on Ubuntu and macOS. Both jobs succeeded on a rerun. The first macOS attempt failed timeout-records-one-failure-and-cleans-children and timeout-leftover-lock-does-not-claim-running (protocol 179/2), while Ubuntu passed. Three local runs that print each condition of that test passed. This is the same kind of intermittent hosted-macOS signal and timeout failure recorded in 2.2.0, and it is still unresolved.
• Locally on macOS (Python 3.12), on the release tree: full selftest.sh 213 passed, 0 failed; protocol 181/0 three times; selftest-followup.py 67/0; every other selftest-*.py and generate.py --selftest exit 0.
• An independent Codex review (gpt-6.1-sol, xhigh) ran five rounds over the change. It found blocking defects in cancellation cleanup, exact marker matching, exit-10 guidance and stale handoffs. All are fixed, and the last round found none.

Not run: live codex or claude validators. Stub and fake-CLI runs show the wiring only, not the quality of a live model's review. The external callers (ACP #1189, Logic Pro MCP #1403) were not exercised.

Install from the tagged source archives using the README. No separate registry package or uploaded binary asset is supplied.

Frontier-simplify 2.2.0

Frontier-simplify 2.2.0 makes follow-up reviews converge on the contract instead of one newly found path per round. The motivating case (#44) was one consumer PR that took fifteen reviews, seven of them in a row on the same receipt invariant.

• Contracts, not cases (#45). A contract defect is reported as one invariant plus a producer/sink table, enumerated once. Follow-ups verify the table row by row; rows found outside it are counted as MISSED.
• Redesign stop (#46). Each contract ends in - invariant: KEY | family N | STATUS | sentence. If the same key stays open in two consecutive recorded reviews, a follow-up returns exit 12 before reserving an attempt and writes REDESIGN-REQUIRED.md from the skill's template. The next review judges the redesign first.
• Witness ledger (#47). Reproductions the reviewer saves in review_evidence/ are preserved outside the checkout. Each one reruns on the next head before any model starts; a failure returns exit 13 without using an attempt. witnesses runs the same check as a preflight. Witnesses can be retired with a reason or disputed.
• Supplementary mode (#48). After the automatic three attempts, --supplementary --granted-by WHO --budget N records an explicit grant. Every round prints the PR's cumulative count and time. A spent budget writes HANDOFF.md asking for one decision. Finding IDs cited from reviews run outside the runner are flagged.
• Pre-existing defects (#49). Base reproduction plus a classification: introduced, inside the change's new contract (offered with a scope-reduction alternative), or outside it and routed separately.

Exits 12 and 13 reserve nothing and launch no reviewer. None of the new lines the host reads can close a finding or authorize a merge, and a review that omits them stops nothing.

Verification

Release commit: 8b71fd32a8d86fa9d3adaf19fa1a0c0a84f10de9 (merge of #50; its tree is identical to the PR head 4dbae41).

• Main CI ran the full offline selftest on Ubuntu and macOS. Both jobs succeeded on a rerun. The first macOS attempt failed sigint-records-one-failure-and-cleans-children and sigint-leftover-lock-does-not-claim-running (protocol 171/2); the identical SIGTERM and timeout cases passed in that same run. The identical tree passed in the PR checks on both platforms, and six local protocol runs, two at a time, gave 173/0 each. The intermittent SIGINT failure on hosted macOS is unresolved.
• On that tree, locally on macOS: full selftest.sh 205 passed, 0 failed (protocol 173/0); the new selftest-followup.py 58/0; every other selftest-*.py and generate.py --selftest exit 0. Each new guard was mutation-checked: removing it turns its tests red.
• The change was reviewed through this runner by codex gpt-6.1-sol in three automatic rounds, using the features it adds. Round 1 found three blockers (a tree-supplied witness helper, phase-1 restarts bypassing the stops, adapter output order) and wrote witnesses for them. The round 2 and 3 prechecks reran those witnesses inside the reviewer's sandbox before any model started. Rounds 2 and 3 recorded all three closed, with no remediation regression and no first-review enumeration misses. The third attempt ended in the designed handoff.

Not run: a live claude executor review, and a measured replay of the #44 estimate (recorded as NOT_RUN). Witnesses are reviewer-written code run by the host; set REVIEW_WITNESS_WRAPPER to run them inside a sandbox.

Install from the tagged source archives using the README. No separate registry package or uploaded binary asset is supplied.

Frontier-simplify 2.1.0

Frontier-simplify 2.1.0 strengthens the optional review runner and its offline validation. Git context and source-index handling now cover literal filenames, alternate indexes and linked worktrees. Prompt input is finite, review inputs are installed without clobbering existing files, review rounds reserve state durably, witness processes are cleaned up, and phase scope and checkout observations are bound to the review evidence. Follow-up guidance and isolated offline evaluation cases are included.

Verification

Release commit: c718df0e23aee37f29b04aa031881514a9ced43c.

• On that commit, REVIEW_SELFTEST_FAST=1 bash skills/frontier-simplify-review/scripts/selftest.sh exited 0: 32 passed, 0 failed.
• Main CI ran env -u REVIEW_SELFTEST_FAST bash skills/frontier-simplify-review/scripts/selftest.sh on Ubuntu 24.04.5 and macOS 26.6.2 arm64. Both jobs succeeded: protocol 173 passed, 0 failed; full selftest 205 passed, 0 failed. The PR checks also passed on both platforms.
• Qualification record separates execution, quality comparison and milestone status.

Live CLI smoke and actual same-input quality comparison were NOT_RUN; offline fixtures do not establish accuracy gains or safe automatic merging. A separate local macOS 26.3 run previously hit a process-start timeout; that host-specific failure remains unresolved despite the hosted macOS full pass.

Install from the tagged source archives using the README. No separate registry package or uploaded binary asset is supplied.

Frontier-simplify 2.0.0 — Fable and Astra

Frontier-simplify is built for frontier coding agents including Fable and Astra (gpt-6-astra): less invented process and bounded repeated code reviews.

Upgrade

This is a major release because repository, marketplace, plugin and public skill paths changed from sol-simplify to frontier-simplify. Remove the old installation, then install the new identity using the updated README. Reinstall any local hook or poller from the new checkout.

Existing ~/.sol-simplify-review history is reused in place, preserving the cumulative three-attempt PR budget. Legacy hooks retain the original user hook; the optional poller installer disables the old job before starting the new one. Historical evidence keeps its original names and model attribution.

Verified

• Full offline selftest: 214 passed, 0 failed; git diff --check clean.
• Fresh Codex marketplace/plugin installation returned version 2.0.0. Installed skill bytes match the checkout, and the installed runner reads existing consumer evidence without adding an attempt.
• Real gpt-6-astra four-arm process-design spot check: 246 / 81 / 152 / 33 lines for plain / Korean one-line / English one-line / full skill. One run per arm; line count is not correctness or safe review convergence. Raw outputs and limitations.
• Fable is an intended use case, not a measured performance claim. No Fable benchmark is reported.

The trusted-local runner preserves cache freshness, read-only status/reporting, finish-time head checks and execution lifecycle handling. Three attempts bound automatic repetition, not defect closure or safe merge. No hosted service or separate registry package is required.