omp-tandem 3.7.0

Changes

• Differing propose now records one pending proposal instead of immediately activating a plan. Active execution, agreements and authorization remain unchanged during negotiation. Proposals carry IDs, base plan revision and card-wide attempt/step impact previews; replacement and proposing-current-plan withdrawal are allowed only before a transition begins.
• Operator CLI transition inspect/begin/resolve/activate/withdraw exposes exact proposal/transition identities, optional revision CAS and whole-command operation fingerprints. Exact retries return their recorded outcome, not permission to repeat effects. Participant views surface exact operator commands and operator_required hints without granting operator authority.
• Begin freezes the proposal and inventories all active/recovery-required attempts, including unchanged steps. Managed disposition requires supervisor stop confirmation; manual disposition requires --confirm-stopped. Operator note/evidence records inspected effects. superseded and abandoned remain distinct; abandonment creates a blocker and pauses the card. Supervisor teardown confirms process stop, not absence of external effects.
• Managed interrupted work is preserved on its recorded composed base with submission-strength byte/ownership checks, or capture failure is retained. Manual saved commits are validated in the pinned repository. Activation resets the active plan to a new draft revision, clears agreements/current acceptance, carries blockers, revokes the grant and preserves existing pauses. Saved continuation is checkpoint, blocked or not_available, never automatic command replay or transferred acceptance. Withdrawal after begin leaves sticky quiescence; emergency operator reconcile remains available.
• Cards record immutable launch-root/scope provenance and initial/latest Git observations at create/propose/claim. Declared ownership and review-context paths reject nested repository/worktree/submodule-content boundaries and symlinks early without opening/searching children. The parent gitlink entry itself is allowed by declared-path validation; pathless non-Git planning is explicitly unverified, while declared paths and claims require Git-root/HEAD.
• Unresolvable submissions fail before intent with the exact commit, pinned repository, bounded Git cause and correct-scope recreation/closure guidance. Source and submitted commit diagnostics are distinct; ancestry, full-hash and ownership validation remain intact.
• Operator cancel closes only disposed cards as cancelled/superseded without fake acceptance, records closure and optional continuation, clears pending negotiations, archives an open disposed transition as cancelled and revokes authorization. Terminal guards prevent reopening execution; history stays readable. Operator link records separately scoped predecessor provenance without transferring permissions, agreements, grants or acceptance. Markdown reports show closure and both link directions.
• EN/RU/ZH README and guides plus both skills document the CLI/MCP workflows and operator boundaries. The two operator requests are preserved verbatim with provenance, separate from implementation decisions. Local package/plugin/marketplace/lock metadata is prepared as minor version 3.7.0 for new commands and changed propose semantics.

Remaining limits

• Release preparation implies no automatic application of the accepted result to a user checkout. Existing tags and archives are unchanged. Independent acceptance, explicit application and publication remain separate decisions.
• CLI/MCP transition evidence uses two manual attempts on disposable Git repositories. Managed confirmation/preservation have store/workspace regressions; the documented end-to-end commands do not run a live-provider replan or a counted external-effect process across managed replan. The pinned OMP verifier uses an isolated localhost model fixture, not user-provider credentials or an external-model evaluation.
• Capture failure yields continuation.status="not_available"; ownership loss or a removed step yields blocked. These continuation outcomes do not automatically block card activation/execution. Operators must inspect preservation failures, remaining work and external effects; there is no generic cross-card importer or automatic rollback/replay.
• Links remain target_verification="not_performed" and reciprocal_link="unverified" even when both sides were recorded. They are operator provenance assertions, not cross-scope read capabilities or proof of a matching target. Fresh agreements and independent acceptance occur only in the successor's scope.
• Boundaries are declared-path and pinned-root checks, not an OS sandbox. Existing managed-snapshot symlink/submodule/filter refusals, unknown-cost stops, independent-review disclosure gates and manual-interaction limitations remain. Historical receipts do not revive fenced authority.
• Compatibility remains version/platform/API-bound; the wire census covers exercised openai-completions, not all providers. Helpers remain disabled with the existing five unsatisfied gates. Prior unexplained cleanup/wake failures remain historical evidence, not explained by later passes.

Cumulative commit accepted by the shared review: f19e74d (replan f88bf79 accepted by OMP; scope d793cf0 and docs f19e74d accepted by Claude). Artifacts: wheel, source archive and SHA256SUMS built by scripts/package.py from the tagged commit.

omp-tandem 3.6.0

Changes

• Bounded shared-work summary/plan/step/full views, paged history and revision/visibility-bound cursors; presentation parameters are separate from mutation identity. Prerequisite hints do not grant authority.
• Runtime identity records loaded distribution/build, checkout observation and registered schema surfaces, separately from execution and model-selection provenance.
• Independent clarification requires a new snapshot; exact unchanged review context and waiver applicability follow the agreed snapshot inputs. Installed-runtime verification exposed and corrected a manual-native claim binding gap in clarification/artifact disclosure guards; manual execution permissions remain unchanged.
• Canonical structured outcomes reject contradictory success/current-check combinations and blocked reports without reasons. Immutable check history preserves original failures and applicable later results.
• Operator-authorized same-claim report-only recovery, explicit application observations/receipts, Markdown work reports and scoped native usage attribution.
• Wake acknowledgments distinguish acknowledged, deferred and not attempted without dispatching work. Verifier preflight, acquisition/setup/probe failure classes, live-parent abort evidence and registered-versus-wire schema census have separate observations.
• Fresh installed-wheel fixture exercises real MCP, pinned OMP RPC/localhost model callbacks and successor CLI with new temporary cards; this is bounded compatibility evidence, not an external-model evaluation.

Remaining limits

• Release preparation implies no automatic application of the accepted result to a user checkout. Existing tags and archives are unchanged.
• Wire census covers exercised openai-completions only. Other APIs remain untested; provider transformations and validation diagnostics are recorded rather than declared universally equivalent. Missing structured finish remains a failure.
• Every task result carries runtime identity; editable/unverifiable exact builds stay unknown. Review-phase refusals are reason codes; (code=cyber_policy) remains a provider-policy refusal, not an automatic retry.
• Non-get tool actions with format=markdown render fenced JSON. Application is not_recorded without evidence; Claude/unattributed costs are unknown and native values may be subtotals.
• One observed TemporaryDirectory cleanup failure in tests/test_review_runs.py passed on reruns but remains unexplained. The historical wake failure also remains unexplained; deterministic interleavings specify behavior, not its original cause.
• Helpers remain disabled with the same five unsatisfied gates in helper compatibility. No new helper execution is enabled.

Integrated commit accepted by the shared review: faffd60. Artifacts: wheel, source archive and SHA256SUMS built by scripts/package.py from the tagged commit.

OMP Tandem 3.5.0 — independent-first shared review and operator grants

OMP Tandem 3.5.0

Features

• Two documented entry paths: review a prepared staged change with tandem_review_run, or run a two-agent shared development task with tandem_work, exact committed submissions and distinct acceptance.
• Operator-pinned managed Claude/OMP selections: --claude-model (documented sonnet default), --omp-model as an alias of --model, and separate selection provenance versus observed actual identity. Run/start cannot override an active grant.
• authorize --preview describes the proposed grant without activation. Explicit --max-attempt-cost-usd is independent of launch count; the default is half the total. Atomic launch-order reservations subtract known spend and active reserves. Grant preview exposes reserve policy and actual permissions.
• Canonical --allow-shell with deprecated --allow-tests alias: both mean arbitrary shell, not a sandboxed test runner.
• Independent-first shared review: immutable complete report, at most one optional compare, then separate exact-submission acceptance/rejection. Author interpretation stays withheld until comparison. Managed reviewers use a server-bound pinned commit reader; no native live-file or arbitrary-shell bypass is advertised.

Fixes

• Unresolved blockers retain identity, origin and resolution history across plan revisions; removed-step blockers move to card scope. Corrective agreement does not enable blocked execution or publication. Only the blocker author or operator can resolve/invalidate with evidence.
• Closed shared-review author-artifact, inferred read/history and claim-response disclosure paths. Shell-granted review records an operator policy blocker before launch; participant-authored lookalike blockers cannot waive it.
• Documentation now states the claim/submit prerequisite: the bound project root must be a Git repository with an immutable HEAD commit, not its parent folder.

Migration

• Legacy grants preserve historical total-budget/max_launches attempt ceilings, allow_tests decoding, active reservations and recovery; ambiguous old OMP selections require explicit reauthorization. Reading/migrating state does not restart execution.
• Legacy review attempts are labelled legacy_disclosure; historical acceptance is not relabelled independent-first. Legacy project import stays copy-only. There is no automatic downgrade or rollback of external effects; use a stopped-state backup procedure before manual restore.

Known limits

• Helpers remain disabled (delegation.available=false). The G0 compatibility map records five unsatisfied gates: child_tools_intersect_parent_restriction, project_agent_definition_cannot_widen_bundled_scout, task_wide_settings_snapshot, parent_stream_usage_scope_exclusive, and no_extra_model_calls_per_spawn. Stages C–F, production scout/sonic delegation and helper cost savings are not released.
• A passing real-OMP verifier means its probes executed, not that unsupported helper gates passed. Startup-auth fallback and post-abort revival remain unverified. Compatibility is version/platform-bound, not a provider or version-range guarantee.
• Stage-confined shell review is unavailable. A shell grant blocks managed review until an evidenced operator decision permits no-shell review; required checks must not be silently waived. Tool restrictions are not an OS sandbox, and manual review cannot erase prior interactive disclosure.
• The recorded tandem_finish schema/description mismatch remains a UX issue; this release does not change that API opportunistically. Structured final reporting is still required.
• test_committed_peer_change_generates_wake_hint_without_dispatch was intermittent during the preceding review: a pending wake remained after get, then the isolated test and full affected suite passed on repeat. Wake acknowledgement is explicitly best-effort under SQLite contention. The original interleaving was not traced; contention is an inference, not a confirmed diagnosis. Integration's first full suite passed without changing this assertion.
• Monetary ceilings use reported/estimated cost, not invoice-hard limits. Unknown cost prevents new launches. No migration, saved card or wake notification grants autonomous execution; applying an accepted result remains a separate operator action. No push, tag, publication or apply is implied by release preparation.

Verification

• Full pytest: 489 passed, 364 subtests (macOS 3.13 and Linux 3.12 CI green on this commit).
• ruff check / format, wheel build, scripts/package.py --check (124 payload files).
• Real Oh My Pi 18.1.13 verifier: 14/14 checks passed; delegation.available=false.

Full specification: docs/spec/helpers-and-shared-development.md. Helper compatibility evidence: docs/helper-compatibility.md.

OMP Tandem 3.4.0 — shared tasks and autonomous collaboration

OMP Tandem 3.4.0

One shared task for Claude and OMP

• tandem_work provides a durable, versioned task card with goals, constraints, an agreed plan, checklist, assigned implementers and distinct reviewers, dependencies, blockers and evidence. Both participants use the same project-scoped state; Markdown is a rendering, not a concurrently overwritten source of truth.
• Plan agreement, execution, submission and acceptance are separate transitions. Optimistic revisions and stable operation IDs reject conflicting changes and deduplicate exact retries. Acceptance references an immutable output and plan version.
• Independent modules can run concurrently in separate Git worktrees. A final integration step combines accepted prerequisites. The original checkout changes only after an explicit operator apply --expected-head.

Explicitly authorized autonomous continuation

• New operator CLI: python -m omp_tandem.work_daemon with authorize, run, start, status, stop, revoke, reconcile, show and apply.
• A separately authorized controller can outlive the interactive client. It starts dedicated Claude/OMP attempts under project, tool, wall-time, launch-count and reported-cost limits. Models cannot grant themselves autonomous authority through MCP.
• Workers acknowledge startup through the shared task. Cooperative blockers preserve source checkpoints; after an evidenced unblock, work continues from the saved edits. Rejected submissions also remain available for targeted correction.
• Explicit pause is sticky. Lost acknowledgements, crashes or uncertain effects do not cause automatic replay. Recovery requires an operator decision after inspecting old execution and side effects.
• Attached clients receive advisory change events; durable state, not notification delivery, controls readiness. Unknown cost pauses autonomous work instead of being counted as zero.

Safety and correctness

• Managed worker capabilities are bound to their task, role and attempt. Reviewers cannot accept their own implementation; managed acceptance is committed only after successful worker completion and snapshot integrity checks.
• Raw committed bytes are preserved despite Git line-ending/ident attributes. Unexpected file edits, unsafe source entries, stale outputs, dependency conflicts and changed apply targets fail closed with retained evidence.
• Corrected edge cases include retired claim receipts, manual submission validation, reconciliation revisions, removed unresolved steps, concurrent stop updates and incomplete native cost accounting.
• Git worktrees are not an OS sandbox. --allow-tests grants arbitrary shell capability. Reported monetary limits are estimates/soft limits, not invoice-level guarantees. Local execution is unavailable while the machine is asleep/offline.

Verification

• Local full suite: 336 tests and 289 subtests passed. Ruff, wheel build and source archive manifest checks passed; installed-wheel CLI startup was exercised.
• Actual Claude Code 2.1.267 and OMP 18.1.13 were exercised through real stdio MCP/native RPC. Only model HTTP replies were supplied by a deterministic localhost fixture; no paid provider calls were made.
• Real-client scenarios covered shared agreement, two parallel modules, cross-review, final integration and explicit apply; detached continuation after client exit; checkpoint preservation and duplicate unblock; and explicit pause stopping both agents without a second dispatch. These are integration checks, not model-quality or comparative benchmark results.
• README and detailed guides updated in English, Russian and Simplified Chinese.

Русское руководство · English guide · 中文指南

Publishing or updating this plugin does not enable unattended work for a project. Autonomous execution requires its own explicit operator grant.

Final release CI

Run 34595972302 passed for f3954e667fd287095e03e208859fbc62b2899890 on Linux / Python 3.12 and macOS / Python 3.13. Each platform passed 336 tests and 289 subtests, real pinned OMP verification, Ruff, wheel building and distribution manifest verification.

Post-release CI clarification

The later tag-triggered run 34596370446 failed on macOS despite the earlier passing branch run at the same commit. Its review lifecycle test temporarily replaced methods on the running controller with assertion-raising mocks; the background driver could hit that test-only replacement and interrupt the run.

Test-only correction 40238ee removes those internal implementation assertions while retaining observable reply, repeated-reply, cancellation and preserved-answer checks. No runtime code, timeout, skip or CI trigger was changed. Corrected CI 34596992268 passed on Linux and macOS: 336 tests and 289 subtests on each, plus real OMP verification, lint, formatting, wheel and manifest checks.

The published v3.4.0 tag and assets remain unchanged; the test correction is on main. The historical failed run is retained, not rerun to manufacture a green result. README and detailed shared-task guides were included in v3.4.0 in English, Russian and Simplified Chinese.

OMP Tandem 3.3.0 — bounded reviews and real verification

OMP Tandem 3.3.0

First useful review, fewer manual steps

• First-run guidance in English, Russian and Simplified Chinese: local project diagnosis, then read-only review of the prepared/staged commit.
• tandem_review_run: stable request identity, bounded start/status/reply/cancel, one independent stage plus at most one author comparison, compact progress and complete final answers. Default 600-second total budget; no automatic edits or shell/test execution.
• Proportional planning: a known local fix gets a brief independent risk check; one comparison, then decide, distinguish by experiment or state the unresolved question. No consensus loops.

Correct capture and cancellation boundaries

• Snapshot capture runs in a separately supervised process with owner/reservation/deadline checks and atomic publication. Late capture cannot claim no_changes or start a native phase.
• A publication gate prevents one snapshot writer from blocking another run's cancellation under the controller guard. Contended stops are explicitly pending until persisted; no false terminal acknowledgement or replay.
• New regression reproduces the old cross-run MCP cancellation timeout and passes after correction; rollback during publication retains snapshot identity without publishing cancelled contents.

Evidence instead of benchmark claims

• Real development case: all three paid episodes retained, including missing-context cancellation and timeout; complete available answers, snapshot hashes, coordinator+peer accounting, and separate runtime follow-up.
• Measured episode wall time totals 598.408s; known cost subtotal USD5.6829274. Full cost and human effort are unknown; prior development/repairs are outside those measurements. This is not a comparative benchmark.
• Real OMP compatibility: official checksum-pinned OMP18.1.13 + SDK revision daf07999c2fee9b22edc7bf8fea1fb6272e0df5e, with a deterministic localhost model fixture and no paid credentials. CI runs the binary, not a fake RPC server.
• Separate four-arm benchmark protocol/schema/offline analyzer; no comparative results or superiority claim.

Verification

• Local full suite: 276 tests and 267 subtests passed; Ruff clean; wheel and deterministic source archive built and verified.
• Actual stdio MCP + official OMP smoke passed: two automatic stages, idempotent start, full 25,200/24,300-character answers, staged-only applicability, stale detection and preserved answers, no native project/shell tools exposed to the review model.
• Real-binary local compatibility passed on Darwin arm64/Python3.13.5: host tools, structured completion, continuation, cancellation, think/analyze/work capabilities. Exact JSON proof is included; this does not certify a broad version range or every provider.
• Two dependency deprecation warnings remain from Authlib; no runtime exceptions suppressed to obtain a pass.

Work mode is not an OS sandbox. A completed review is not proof that code is correct. Channels/hooks remain optional; bounded polling remains supported.

Русский: понятный первый запуск, компактное независимое ревью, ограниченное обсуждение, исправленные deadline/отмена и честный разбор трёх попыток. README и подробное руководство обновлены.

中文：更新首次使用流程、紧凑独立审查、有限轮次讨论、捕获期限与取消修复，以及保留三次尝试的真实案例。中文 README 和完整指南已更新。

Final cross-platform acceptance

• CI run 34459190147 passed for release commit 5f1136b5a32cee5314c120a68d9ff7662162ba37 on Linux/Python3.12 and macOS/Python3.13. Both jobs passed 277 tests and 267 subtests, the actual pinned OMP verification, Ruff, wheel build and distribution manifest check.
• CI follow-up corrected a concurrent-close race and foreign-owner validation order, and synchronized deadline regressions independently of interpreter startup speed. Production timeout limits were not increased.

v3.2.0 — Plan first and review staged material

Plan first, choose the reviewed material explicitly

• The tandem skill and coordinator instructions require peer planning before nontrivial implementation: establish the need and criteria, obtain independent framing, compare the coordinator proposal, record concrete file ownership/order/checks, implement, and cross-check. Only mechanical changes or unchanged explicitly user-approved plans use a shortened path. This is an instruction policy, not a new API permission gate.
• tandem_review now accepts request.source="worktree"|"staged", defaulting to worktree. Staged snapshots select index/base differences and derive selected bytes, file modes, change classification and diff from the index without reading working files. Later unstaged/untracked changes cannot leak into the review.
• Staged applicability tracks the selected index entries; subsequent unstaged edits leave the snapshot current, while selected index changes make it stale. Legacy source-less snapshots retain their saved bytes and hashes and are interpreted as worktree.
• Profile selection now explicitly explains that balanced and deep both default to high thinking; deep increases the deadline from 1800 to 3600 seconds, not reasoning depth. tandem_scope.execution_profiles exposes a catalog of defaults; overrides/effective/actual settings and permissions are unchanged.
• Updated English, Russian and Simplified Chinese guides and landing pages. The MCP tool set remains 18 tools.

Verification

• 228 tests and 236 subtests passed locally; Ruff, wheel and distribution-manifest checks passed.
• Actual OMP independent planning preceded edits; a second turn compared the coordinator proposal. The completed saved implementation then received an independent peer cross-check and comparison against the plan, with no concrete correctness findings.
• Real MCP/Astra smoke read INDEX_ONLY from a staged snapshot despite a different working-tree version. Unstaged mutation preserved the fingerprint and applicability; a later git add made the old snapshot stale without changing its saved answer/material.
• Smoke checked profile discovery, default/effective separation, explicit override precedence and catalog isolation.

Staged snapshots require Git and retain the existing capture limits/conflict/submodule restrictions. Newly staged paths outside a saved bundle are not implicitly reviewed; recapture for a changed complete commit candidate. Reload the plugin/client after updating to load the new skill and runtime.

CI passed on Ubuntu/Python 3.12 and macOS/Python 3.13: https://github.com/Flyozzzz/omp-tandem-public/actions/runs/34442314789

v3.1.0 — Reliable delivery and version-bound review

Reliable delivery and version-bound review

• Productized Claude asyncRewake watchdog: bounded lifetime, authenticated live arm, session/task/generation/incarnation ownership, duplicate suppression, and real wake acknowledgment. Without verified current coverage, use bounded polling even when push is confirmed.
• Added durable result receipts: claim once before result-driven effects, complete after handling, retain uncertain claims across interruption. Reads and notification acknowledgments do not authorize replay or grant new permissions.
• Added immutable review bundles with captured selected/base/staged bytes, commit identity and hashes, derived diff, supplied checks, separate author rationale, and saved/observed/external boundaries. Results report current/stale/previous/unknown applicability.
• Snapshot reviews use saved-material host tools in think mode. Comparison reveals the author proposal only after a completed independent assessment of that snapshot in the same conversation.
• Added current-client diagnostics: local by default; explicit live check runs one short provider task, reports actual model/authentication and delivery, and reuses its task ID for follow-up.
• Added quick/balanced/deep computation profiles and explicit overrides independent of permissions. Ordinary results include actual settings, duration, token categories and available native-reported cost for the task and whole conversation, retaining unknown/partial values.
• Added stable findings with numbered lookup, append-only evidence history, optimistic revisions, separate validity/fix state, and successful snapshot-bound verification. Worker report ingestion is atomic and idempotent.
• Updated English, Russian and Simplified Chinese guides and both skills. MCP surface now contains 18 tools.

Verification

• 218 tests and 236 subtests passed locally; Ruff, wheel build and distribution manifest verified.
• Real OMP/Astra smoke covered immutable reads despite working-tree mutation, independent/comparison turns, preserved first answer, fix verification on a new snapshot, usage accounting and live diagnostics.
• Actual Claude Code 2.1.266 with shipped hook commands recovered the original task after an injected send exception and after notifications were dropped following real transport writes. The first control recheck was still running; waiting continued without task restart. Receipt claim responses were true/false, with one completed receipt.
• Corrected the watchdog decoder for the actual Claude JSON-string hook payload, retained as regression coverage.

Boundaries

Snapshots certify selected saved material, not the whole environment. Capture is bounded and rejects selected submodules/unmerged index entries. Supplied test output remains a claim, native cost is not an invoice, and receipts cannot guarantee exactly-once arbitrary external actions. Reload the plugin/client after updating to load the new hooks and runtime.

CI passed on Ubuntu/Python 3.12 and macOS/Python 3.13: https://github.com/Flyozzzz/omp-tandem-public/actions/runs/34375113672

v3.0.2 — Delivery-aware and independent-first collaboration

Delivery-aware collaboration

• Shared collaboration guidance plus separate polling and confirmed-push procedures; no changes to the 14 MCP tools.
• Scope/task responses provide current delivery_instructions. Push requires real probe receipt; transport fallback restores bounded polling and consistent next_action values.
• One polling task uses tandem_result(wait_seconds=25); several use tandem_wait then fetch ready results. Confirmed push uses events without a polling loop.
• Consequential consultations/reviews use independent-first framing: original task, facts, constraints and code first; coordinator proposal and arguments in a follow-up. Prior exposure must be acknowledged; simple execution need not use two stages.
• Updated the collaboration skill and English, Russian and Simplified Chinese guides.

Verification

• 166 tests and 214 subtests passed on macOS; Ruff checks and wheel/distribution validation passed.
• Real stdio protocol tests cover forced polling and confirmed push; regression covers rejected receipt and transport fallback.
• Real OMP consultation over stdio completed both independent-first stages; the original assessment remained unchanged in task history. Initialization delivered distinct forced-poll/automatic guidance within the 2048-character limit.
• GitHub CI passed on Ubuntu/Python 3.12 and macOS/Python 3.13: https://github.com/Flyozzzz/omp-tandem-public/actions/runs/34364336434

v3.0.1 — MIT, clean repository, focused services and CI

MIT-licensed OMP Tandem snapshot for future open-source distribution.

• MIT LICENSE included in the source ZIP, wheel, package metadata and plugin manifests.
• Concise English, Russian and Simplified Chinese landing READMEs; full reference guides moved to docs/.
• Contributor and security policies, issue forms, pull-request template, repository topics and updated installation URLs.
• bridge.py reduced from1210 to120 lines. Real services now own persistence/recovery, task admission, native execution, questions/reports, contracts and result projection; public MCP behavior retained.
• pytest is an explicit development dependency. uv run pytest -q runs in a fresh project environment without relying on a global pytest or PYTHONPATH workaround.
• Pinned, least-privilege GitHub Actions checks pytest, Ruff, wheel building and the distribution manifest on Ubuntu/Python3.12 and macOS/Python3.13.
• Hosted macOS testing exposed loopback HTTP startup's unnecessary reverse-DNS dependency. It is removed and covered by a regression without weakening timeouts or disabling tests.

Verification:165 tests and214 subtests passed locally; both hosted CI matrix jobs passed. Native OMP task execution and same-conversation continuation passed after the service refactor. Wheel metadata reports MIT and contains LICENSE; source files and release packaging were checked.

This repository contains only a new reviewed snapshot and its CI hardening commit; no commits, tags or releases from the former private repository were imported. The old repository was not force-pushed or made public. Both repositories remain private until an explicit owner decision. History cleanup cannot erase third-party clones or caches; exposed credentials must still be rotated.