Conversation
Bring credentials, requests, operations, protobuf helpers, and generated clients into product parity while retaining JS runtime mechanisms. Include the compatibility matrix and validation records. Written by Codex (gpt-6/Codex) on behalf of Daniel Drizhuk
Correct setup instructions, migration guidance, and public API comments. Require public comment coverage and STE in repository agent rules. Written by Codex (gpt-6/Codex) on behalf of Daniel Drizhuk
|
Compatibility matrix for SDK behavior changes Reports moved from repository files to this PR discussion. The 59 contracts below use public GoSDK Accepted choices (the user selected all recommendations):
Compatibility matrix — all 59 contracts
“Implemented” records a code change. “Reviewed” records retained behavior checked against source and the local suite. Language-specific results preserve JS mechanisms. No decision remains pending. Written by Codex (gpt-6/Codex) |
|
Behavior-change audit and validation record The audit before this documentation follow-up completed 34 fresh primary passes, two separate SDK parity audits, and nine independent Opus cross-reviews. Incomplete review attempts are excluded. The final primary found no actionable findings. The final Coverage included configuration, credentials/acquisition/recovery, authorization, transport, requests, operations/progress, protobuf wire/JSON/Any/extensions/native values, masks/traversal, pagination, generated services, diagnostics, and shutdown. The final primary compared public layers against both reference SDKs. The final Opus comparison sampled those layers; Python token implementations and Go mask filter/traverse were not separately compared in that cross-review. Validation before the documentation follow-up:
Limits: generated message fields were sampled; third-party runtime internals were not audited exhaustively. No production calls or live identity-provider login were run locally. Dayjs keeps millisecond precision. Private-only SDK features are outside this public-source parity scope. Public schema inventory{
"publicMethods": 481,
"generatedMethods": 481,
"publicMessages": 1219,
"generatedPublicMessages": 1219,
"missingMethods": [],
"missingMessages": [],
"extraMethods": [],
"schemaRevision": "43cab7c03f9b162b2e524d03b7c47fd74451fcd9"
}Documentation follow-up review: five completed fresh primary passes and four independent The final Documentation validation passed with zero missing-documentation validation warnings. The 135 existing duplicate-declaration conversion warnings remain. README examples type-check against the built package; touched-file lint, formatting and whitespace checks pass. All ten changed TypeScript files emit byte-identical JavaScript when comments are removed. The five external report backups match the deleted blobs. Original workspace staging was preserved throughout review and publication. Follow-up commit: Written by Codex (gpt-6/Codex) |
The JS SDK diverged from public Go/Python behavior across credentials, requests, operations, and protobuf helpers. This change aligns the product contracts while retaining JS events, promises, native values, and client ownership. Failed operation waits now reject with an inspectable
OperationError; automatic credential discovery remains opt-in.NEBIUS_IAM_TOKEN, align CLI credential priority, and reject invalid explicit credentials.toJSONoutput and correct wire merging, wire-tag validation, extensions, and unsigned values.AGENTS.md.The candidate is based on latest JS main
644aa844, public API43cab7c, GoSDK83393ae1, and PySDK94511726. All 481 public RPCs and 1,219 messages are generated, with no missing entries.Validation: 430 tests pass across 34 suites, including 50 generator tests and both rebuilt-package tests. The full CJS/ESM build, runtime lint, scoped generator lint, changed-file formatting, and whitespace checks pass. Generator reference tests use local ts-proto 2.6.1; the repository configuration is restored. Production E2E calls and live identity-provider login were not run locally.
Final fresh primary review found no actionable findings. The independent
claude-opus-5-5cross-review considers the code merge-ready; its latent helper-name collision concern has no match in the public schema. Remaining suggestions were style nits. Detailed coverage and limits are recorded in the PR audit comment.Documentation follow-up: strict documentation validation, README example type-checking, touched-file lint and formatting pass. The final primary review is clean; the final Opus review’s remaining contributor command clarification was applied. No findings remain unresolved.
Compatibility matrix: #285 (comment)
Audit and verification: #285 (comment)
Written by Codex (gpt-6/Codex)
on behalf of Daniel Drizhuk