{{ message }}
Repository navigation
Tags: veryfront/veryfront-code
Tags
Prepare stable 0.1.1283 for qualified runtime repairs (#5346) * Prepare one coherent stable artifact identity for qualified runtime repairs Promote the merged runtime, management safety, and MCP fixes to one stable version before producing fresh distribution and native proof artifacts. The generated hydration runtime changes only its embedded version. Constraint: Publication requires separate explicit approval for the final release PR Rejected: Reuse earlier candidate artifacts | main gained runtime input changes Tested: Pinned supported generation; exact four-file version diff; source review clear; git diff --check Not-tested: Fresh distribution, packed consumers, and native functional proofs are pending * Keep embedded dependencies reachable beneath compiled project directories Compiled binaries can retain a project directory below the VFS extraction root. Preserve that directory in both esbuild extractors and framework source resolution so native bundling and embedded worker sources remain reachable. Tested: Actual compiled CLI failure reproduced; five causal path regressions; pinned Deno 2.9.7 suite with 19 tests and 121 steps; typecheck and lint; uncommitted review clear. Not-tested: Rebuilt compiled executable and refreshed package artifacts remain to be qualified. Scope-risk: moderate * Preserve compiled extension roots without duplicating the core parser The esbuild extension knows its location relative to the dependency root. Resolve that anchor directly and retain generic fallback layouts, preserving project prefixes embedded by Deno while keeping independent package boundaries. Constraint: Sonar rejects duplicated new lines; the extension remains independently packaged. Rejected: Cross-package internal imports or quality-gate exclusions | neither is needed for a local path lookup. Tested: Pinned Deno 2.9.7 affected suites: 19 tests and 123 steps pass; typecheck, lint and diff checks pass; new regression fails before the change. Not-tested: Fresh compiled and packed artifacts and current-head hosted CI remain pending. Scope-risk: narrow
Prepare the native recovery fixes for stable consumers (#5329) Advance the synchronized package and runtime version after the qualified native source MIME, response-bound, and branch-history repairs merged. Regenerate hydration with the pinned toolchain; all four edits only replace 0.1.1281 with 0.1.1282. Constraint: Publication and release-PR merge require separate explicit approval. Tested: Release version tests; exact four-file substitution proof; source-only uncommitted review. Not-tested: Final packed consumer qualification is in progress. Scope-risk: narrow Confidence: high
Prepare stable release qualification for the native OKF runtime (#5315) Align the framework manifest, workspace identity, shared runtime version and generated browser bundle with 0.1.1281 so the existing stable CI path can qualify the integrated native OKF changes. Constraint: Stable publication remains subject to the normal protected CI release gates. Rejected: Promote the RC dist-tag or bypass the email harness stable contract | neither qualifies an immutable stable release Confidence: high Scope-risk: narrow Tested: Pinned Deno generation and manifest checks; 14 release and CI tests with 91 steps; read-only uncommitted review Not-tested: Stable registry publication and staging acceptance
Preserve managed recovery authority and prepare SDK 0.1.1280 (#5260) * Preserve recovery authority across untrusted runtime and producer lifetimes Capture mutable Promise, Map and filesystem operations on managed recovery paths, keep leases through producer settlement, separate public parsing from provider provenance and validate initial checkpoints before publishing. Update custom test producers to report actual completion without relaxing unknown-producer fencing. Related: SDK #5213 review findings 4304903774, 4304903777, 4304903781, 4304903784 Tested: Pinned Deno 2.9.7 production check; 170 integration tests and 137 steps; six explicit resource-sanitized deferred producer cases; format and lint Tested: Required uncommitted review; its sibling test completion finding addressed with red/green coverage Not-tested: Fresh packed crash qualification, hosted CI and staging acceptance remain Confidence: high Scope-risk: moderate * Give the recovery authority follow-up a publishable release identity Prepare SDK 0.1.1280 through the existing version helpers and generator. Keep package metadata, workspace lock, runtime constant and embedded hydration runtime synchronized so the normal main release workflow publishes the repaired cohort. Tested: Required uncommitted review; its runtime-version finding addressed through generated artifacts Tested: Supported generation task, release-version tests, whitespace checks Not-tested: Public publication and staging acceptance remain Scope-risk: narrow * Keep checkpoint validation aligned with the required formatter Tested: Required uncommitted review clear; all 22 focused writer tests; full formatting and lint checks Scope-risk: narrow * Release failed framework startups without weakening producer fences The built-in runtime could reject transport, model-readiness or memory preflight before reporting its producer lifetime. Managed recovery then treated a known empty startup as an unknown producer and retained session capacity. Reserve completion before preflight and settle it on failure only while no producer has started. Mark runtime invocation at the actual dispatch boundary. Rejected: Stop renewal whenever completion is missing | an invoked override or failed sandbox cleanup can still leave effects running Confidence: high Scope-risk: moderate Tested: Original real-owner startup regression fails before repair; focused cancellation/recovery tests and positive invocation-boundary cases pass; production typecheck, formatting, lint and required review Not-tested: Final normal publication and staging lifecycle acceptance remain
fix(agent): expose optional bootstrap tools within authorized grants (#… …5200) * Keep essential tool schemas available without requiring unauthorized capabilities Add optional startup schema hints that intersect the final authorized catalog. Required bootstrap tools stay strict; optional hints yield to mandatory schemas and leave deferred discovery capacity. Carry the contract through hosted, project, service, and Markdown agents. Prepare the normal 0.1.1278 candidate. Constraint: Read-only agents must start when form or file tools are unavailable Rejected: Make form and file tools mandatory | fails restricted authorized catalogs Directive: Optional hints never grant tools or displace mandatory bootstrap schemas Confidence: high Scope-risk: moderate Tested: Deno 2.9.7; 230 focused tests, 82 exposure/provider tests, release-version tests, types, lint, format, generated manifests, uncommitted review Not-tested: Published artifact consumers and unchanged deployed staging form acceptance remain release gates * Honor optional startup hints without adding unavailable tool dependencies Prioritize authorized optional schemas before eager provider capping, after existing mandatory and local priorities. Do not enumerate tools for a model without tool calling solely because it has optional hints. Keep required bootstrap validation and startup execution loading unchanged. Constraint: Optional bootstrap must preserve existing authorization and provider limits Rejected: Load the catalog for every optional hint | creates avoidable MCP failures for non-tool models Confidence: high Scope-risk: narrow Tested: Both review defects reproduced red; Deno 2.9.7 expanded 313 tests and 223 steps, types, lint, format, uncommitted and fresh-base reviews Not-tested: New exact-head full hook, packed artifacts, consumer cohort, and unchanged deployed staging acceptance remain gates
Preserve canonical event history through strict SDK reads (#5107) * Expose lossless canonical event reads through the stable SDK Validate canonical-v1 pages and stream members against the normative schemas and payload semantics. Preserve mixed legacy rows and owner identity through the typed runs facade. Prepare stable 0.1.1273 and resolve executor test React imports when npm build assets are absent. Constraint: Preserve normative resource bytes and existing pilot defaults Rejected: Drop legacy fallback members | would hide authorized retained history Scope-risk: moderate Directive: Keep canonical resource validation strict and preserve mixed fallback payloads Not-tested: Published stable artifact and final deployed consumer acceptance remain pending Tested: Full pinned Deno 2.7.7 test task, Node executor groups, format, lint, types, branch review, all 31 stable package builds and strict NodeNext declarations * fix(events): apply canonical read validation budgets per member * chore(events): preserve normative semantic resource bytes * fix(events): isolate canonical schemas and fail closed on stream bytes and cancellation * fix(events): defer canonical parser loading and reject malformed HTTP error bytes * fix(runs): verify requested canonical event identity * docs(runs): use the configured SDK in canonical examples * fix(runs): scope canonical event reads to the requested run * fix(runs): enforce requested run across canonical list and stream reads * fix(runs): reject regressing durable stream checkpoints * fix(runs): validate paginated event cursor ordering * Preserve HTTP retry classification when canonical error bodies are malformed Keep deterministic client failures from being retried or losing their response status. Decode failures now produce a sanitized status-bearing error while valid canonical problems and aborts retain their existing behavior. Document the supported Google structured-output capture profile. Constraint: Unvalidated provider response bodies must not enter error details Rejected: Retry every decoding exception | deterministic 4xx responses repeat and lose HTTP classification Confidence: high Scope-risk: narrow Tested: 65 HTTP tests including 45 status, malformed-body, retry and privacy cases; focused types, lint and format; uncommitted review clear Not-tested: Deployed final Events cohort and stable publication remain pending Related: #5107 (comment) * Publish canonical Events support in the next stable SDK Prepare stable 0.1.1274 in the existing Events PR so the normal main publication workflow can release the qualified package cohort. Constraint: Preserve immutable canonical resources and use the normal protected publication workflow. Tested: Supported generation and manifest check; release tests; hydration tests; uncommitted review clear. Not-tested: Final-head full required checks and stable artifact build are pending. Scope-risk: narrow * Keep native watcher measurement outside competing test writes Run the existing native watcher tests in the existing serial group and retain their unit and coverage ownership. The planner regression checks exact ownership and process isolation. Constraint: Preserve native assertions and the existing timeout; do not skip or retry failures. Tested: Full serial suite 13 tests and 479 steps; planner and inventory 8 groups and 55 steps; targeted types, lint and formatting; uncommitted review clear. Not-tested: Final-head full pre-push checks and CI remain pending. Scope-risk: narrow * Accept permanent event store identities across canonical reads Genuine API pages and SSE use an immutable physical-store source base. Validate that opaque base while preserving exact producing-run and stream components and rejecting URL normalization aliases. Constraint: Canonical normative resources and permanent store identities remain unchanged Rejected: Remove store prefixes from API sources | breaks physical-store identity Confidence: high Scope-risk: narrow Tested: Real API page and 11 SSE members; focused parser regressions; clear uncommitted review Not-tested: Final published artifact and deployed acceptance remain pending * Keep supported diagnostic event history readable within bounded budgets Canonical model-input events can exceed ordinary schema-adapter limits. Request per-compiled-validator UTF-8 budgets for canonical members, keep default validators isolated, and bound canonical SSE framing consistently. Constraint: Immutable event schemas and default adapter limits remain unchanged Rejected: Raise global validator limits | broadens unrelated callers Confidence: high Scope-risk: moderate Tested: 74 focused tests and 58 steps, real large HTTP and SSE reads, oversized rejection, types, lint, format and clear uncommitted review Not-tested: Final full suite, published artifacts and deployed acceptance remain pending * Preserve every descendant event across physical store boundaries Select an additive descendant pagination profile explicitly and retain the API's opaque continuation token. Validate the shared strict page and cursor contracts without changing immutable single-run decimal pagination. Equal event IDs from different producing runs remain distinct, while numeric after_event_id stays exclusive and malformed continuations fail closed. Constraint: Existing canonical-v1 resources and single-run cursor semantics remain unchanged Rejected: Reuse one numeric cross-store cursor | equal physical IDs can silently skip history Directive: Server owns cursor authorization and full physical-store tuple ordering Confidence: high Scope-risk: moderate Tested: 73 focused parser, HTTP/SSE and export tests; full unit suite; typecheck; lint; format; required reviews Tested: Actual 31-package npm build and packs; API codec vector with emitted SDK and explicit selector Tested: Studio 773 tests and Agent 433 unit plus 393 eval tests against the exact packed cohort Not-tested: Final deployed staging and production strict Events acceptance remains pending * Keep public event references aligned with the descendant read contract Regenerate the public event and Runs references with the pinned Deno version after adding explicit descendant-page decoding and cursor types. This addresses the exact-head CI reference freshness failure without changing shipped sources. Constraint: Preserve the qualified runtime and immutable 31-package artifact cohort Scope-risk: narrow Confidence: high Tested: Pinned API reference regeneration and freshness check; uncommitted review; diff whitespace check Tested: All 31 real package archives exclude both changed generated reference paths Not-tested: Server-side Sonar reliability gate still requires classification of immutable S3403 finding * Make the hosted deadline regression independent of scheduler load The pre-push parallel suite correctly timed out the test's ordinary retry when CPU scheduling exceeded its 20 ms wall-clock budget. Advance the existing test clock explicitly for stalled resolutions, then let the normal retry complete without artificial time advancement. Preserve the timeout, ingress permit and lookup/dispatch assertions without changing production deadlines. Constraint: Keep the qualified production runtime and packed cohort unchanged Rejected: Increase the timeout or retry the failed suite blindly | either hides scheduler-sensitive test behavior Confidence: high Scope-risk: narrow Tested: Exact hosted HTTP resolver suite, 22 steps; direct test typecheck; uncommitted review * Reject canonical history that violates producer source semantics Validate declared UTF-16, trim and JSON byte constraints with one union-aware traversal shared by synchronous and asynchronous adapters. Reject aliased physical-store prefixes without changing frozen contract resources. Constraint: Preserve immutable canonical resources and opaque JSON payloads Tested: 69 focused tests and 29 steps; format, lint; uncommitted review clear Not-tested: New packed cohort and exact-head CI remain required Scope-risk: moderate * Keep canonical validation regressions type checked Narrow synchronous adapter results, required union branches and structured HTTP errors explicitly without weakening regression assertions. Tested: Direct Deno types; 68 tests and 26 steps; uncommitted review clear Scope-risk: narrow * Validate canonical schema input at the boundary Accept unknown schema input and narrow it before object traversal so the source refinement walker satisfies the anti-slop boundary rule. Change-Type: fix Tested: source-constraints.test.ts; deno check; lint:anti-slop Co-Authored-By: Codex <noreply@openai.com> * Keep arbitrary canonical fixtures typecheckable The equality helper infers its comparison type from the parsed result, while these fixtures deliberately retain untrusted input shapes. Compare them structurally as unknown without asserting that the input is already canonical. Confidence: high Scope-risk: narrow Tested: Frozen deno check of exact failing entrypoint; refinement suite 3 steps; uncommitted review * Keep imported canonical regression fixtures type safe Preserve the content type as the required literal so equality assertions accept the parsed canonical member without widening fixture metadata. Tested: Direct regression fixture typecheck Scope-risk: narrow * Keep bounded schema snapshots within provider latency gates Count UTF-8 bytes without allocating an encoded array per string or key. Preserve captured intrinsic behavior and every existing snapshot limit; exercise all UTF-16 code units and deterministic surrogate combinations against TextEncoder. Constraint: Preserve the unchanged 750 ms required-list regression gate and data-only JSON boundaries. Tested: Provider compatibility and JSON security suite 5 tests/118 steps; final JSON adapters 3 tests/32 steps; exhaustive Unicode and intrinsic replacement checks; lint; uncommitted review clear. Confidence: high Scope-risk: narrow * Reserve an unpublished version for the Events runtime cohort Version 1275 was published independently from main without the canonical Events client. Advance the candidate to verified-unused 1276 and regenerate its embedded hydration version using the standard generator. Constraint: Published npm versions are immutable; never replace the 1275 cohort. Tested: Standard generation, exact generated version-only diff, focused fmt/lint, release helpers 1 test and 13 steps, uncommitted review clear. Scope-risk: narrow Confidence: high * Make shared graph-limit coverage independent of request scheduling Await the unrestricted render entry read before joining the constrained follower. Asynchronous key preparation otherwise lets the constrained request lead and correctly causes an extra retry read. Preserve graph-limit rejection and the exact shared-source read assertion. Confidence: high Scope-risk: narrow Tested: Deterministic reversed-leader red; official module-fetcher suite 90 steps; format and lint; uncommitted review Not-tested: Full hook on this final commit * Keep completed bootstrap writers out of warm HTTP requests Long-lived executor channels retained the capture context of their installation task. Later anonymous requests inherited its revoked writer and failed before model inference. Clear both observer lanes at the independent HTTP boundary and scope fresh request capture normally through lazy response reads. Constraint: Writer revocation must remain enforced after its owning invocation finishes Rejected: Keep the bootstrap writer alive | grants unrelated requests stale authority Confidence: high Scope-risk: moderate Tested: Real warm anonymous transport red reproduction; 46 focused steps on Deno 2.7.12 and 2.9.7; 467 task and capture preservation steps; typecheck; uncommitted review Not-tested: Deployment of this candidate to staging remains pending * Keep global intrinsic mutation out of the hermetic unit suite Preserve the UTF-8 intrinsic replacement regression in integration coverage, while retaining all pure snapshot cases in the unit suite. This satisfies the semantic boundary audit without adding migration debt. Confidence: high Scope-risk: narrow Tested: Schema unit and integration cases; semantic boundary audit; formatting; lint; uncommitted review Directive: Keep process-global intrinsic replacements in integration coverage --------- Co-authored-by: Codex <noreply@openai.com>
PreviousNext