Repository navigation
chore: complete Python npm and NuGet registry metadata - #306
Merged
Merged
Conversation
Releasing was by hand: bump the version, edit the CHANGELOG, merge, then twine/npm publish from a laptop, with a tag and a GitHub release only now and then (traceai-livekit-v0.1.2 is the one package tag). traceAI-crewai 0.1.9 was bumped but never published, traceAI-a2a and traceAI-minimax never at all, and 17 packages report a __version__ that differs from their pyproject.toml. release.yml runs on every push to main. A planner (.github/release/ release_plan.py, with tests) lists the 50 Python and 41 npm packages, asks PyPI and npm which versions exist, and hands the unpublished ones to two reusable workflows, the core packages first. Per package: build, twine check or pnpm pack, install in a clean environment and import (CommonJS and ESM for npm), check the installed version, publish (PyPI trusted publishing or PYPI_API_TOKEN; npm with provenance), wait until the registry serves it, tag `<name>-v<version>` / `npm-<name>-v<version>`, and create the GitHub release from the package's CHANGELOG section (.github/release/release_notes.py). Until an owner adds the credentials (PYPI_TRUSTED_PUBLISHING=true with a trusted publisher per project, or PYPI_API_TOKEN; NPM_TOKEN) every run builds, checks and uploads the artifacts, then warns that nothing was published. Pull requests that touch a version only check and show the plan. workflow_dispatch takes a package list and a dry-run flag. `traceai` (python/traceai) is on the skip list: its PyPI name belongs to another publisher. The 17 __version__ strings now match their pyproject.toml, and the planner's --check (run in the workflow on every pull request) keeps them that way. CONTRIBUTING.md documents the process. Java and C# are not covered. Verified locally: planner tests, --check on the repository, a real plan (crewai 0.1.9, a2a 0.1.0 and minimax 0.1.0 would publish), the Python build + twine check + wheel smoke on traceAI-openai, the pnpm pack + tarball smoke on @traceai/openai, actionlint.
…e caller jobs A caller job must grant what a reusable workflow's jobs request, or the run fails at startup: contents: write for the tag and the GitHub release, id-token: write for PyPI trusted publishing and npm provenance.
…branch Co-Authored-By: Claude Code <noreply@anthropic.com>
Build on Tanmay’s #300 publisher with one release-please PR, publisher-owned tags, secure top-level PyPI OIDC, Maven Central, Go and NuGet. Cover every manifest and test root-core version isolation and cross-language updates. Co-Authored-By: Claude Code <noreply@anthropic.com>
Align the deployment environment with the actual release profile in #302; no local GnuPG keyring is needed. Co-Authored-By: Claude Code <noreply@anthropic.com>
Audit all Python and npm manifests against registry upload timestamps and apply one-time patches to 38 packages whose same-version changes would otherwise be stranded. Keep metadata and dependency ranges unchanged, validate bootstrap version floors, and prove all Java reactor parents including the companion BOM bump together. Co-Authored-By: Claude Code <noreply@anthropic.com>
Add project links, discovery metadata, missing package readmes, and NuGet assets without changing package identities, versions, or dependency ranges. Co-Authored-By: Claude Code <noreply@anthropic.com>
Distinguish verified CommonJS setup from packages blocked by pre-existing entry-point paths. Keep runtime fields unchanged in the registry metadata lane. Co-Authored-By: Claude Code <noreply@anthropic.com>
…s at the emitted build Repair all 18 tarballs with missing declared targets. Export their emitted CommonJS implementation for both require and native import, preserving named exports; drop invalid module/esnext declarations. Version, dependencies and build scripts are unchanged. Co-Authored-By: Claude Code <noreply@anthropic.com>
Verify documentation homepages independently, drop Python 3.14/3.15 claims blocked by the core dependency, initialize async context, and disclose provider runtime limits with tested ESM application-span fallbacks. Co-Authored-By: Claude Code <noreply@anthropic.com>
Use explicit non-recursive core version updaters so manifest-only dependency fixes release safely. Stage exact transitive workspace tarballs for clean-install dry runs, then publish all npm packages sequentially in dependency order. Cover backdated branch merges, align the real strategy test with the deployed action, and fix bootstrap changelog whitespace. Co-Authored-By: Claude Code <noreply@anthropic.com>
Co-Authored-By: Claude Code <noreply@anthropic.com> # Conflicts: # python/frameworks/llama_index/pyproject.toml # python/frameworks/smolagents/pyproject.toml
Synchronize the 18 repaired manifests, runtime versions, changelogs, and release-please inventory. Preserve their metadata and dependency fields. Record explicit version floors without rewriting the bootstrap audit. This avoids publisher tag reconciliation stranding the existing 0.1.0 fixes. Co-Authored-By: Claude Code <noreply@anthropic.com>
PyPI accepts only one pending trusted publisher per repository, workflow and environment, so the 19 first-time projects cannot share one. The publish job now runs in pypi-<project name>; publishers on existing projects keep a blank environment, which accepts it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ap token PyPI allows only three pending trusted publishers per organization, so 16 first-time projects cannot be created through trusted publishing. They are listed in config.json pypi_bootstrap_token; the plan marks a listed package that has never been published, and only those uploads use the PYPI_BOOTSTRAP_TOKEN secret. Every other upload keeps trusted publishing. RELEASING.md records the cleanup after the first release. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The npm publish job installs npm 11.21.1 (trusted publishing needs 11.5.1 or later). npm tries OIDC first and falls back to NODE_AUTH_TOKEN, so the first release keeps using NPM_TOKEN and later releases switch to trusted publishing once per-package publishers exist. RELEASING.md records the migration, including npm's 2-day expiry on unused publisher configs and the workflow_call rule (configure release.yml). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 31, 2026
Accept an explicit trusted-publishing opt-in alongside token authentication. Fail real main releases with pending npm work when neither path is configured, and cover the workflow shell with offline credential-gate regressions. Co-Authored-By: Claude Code <noreply@anthropic.com>
Preserve the latest PyPI bootstrap/environment, npm tooling and OIDC credential-gate fixes from PR #303. Retain all eighteen 0.1.1 release-manifest entries already synchronized by PR #306; combined runtime/manifest, bootstrap, planner and strategy regressions pass offline. Co-Authored-By: Claude Code <noreply@anthropic.com>
Resolve traceai_autogen/__init__.py: dev (#299) and this PR both set 0.2.1; keep the x-release-please-version marker release-please needs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The test hardcoded 0.1.0 and fails now that release automation bumps the package to 0.1.1; it was skipped while cerebras was on the known-broken list (#293 removed it). Compare with installed metadata instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Carry #303's cerebras version-test fix. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Resolve against core 1.2.2 (#313), MCP 0.1.3 (#311) and Vertex AI (#312): - version.py: keep 1.2.2 with the x-release-please-version marker. - mcp CHANGELOG: keep the real 0.1.3 fixes from #311 instead of the bootstrap placeholder for the same version. - vertexai CHANGELOG: record #312's fixes under the 0.1.11 release. - manifest: python (fi-instrumentation-otel) 1.2.1 -> 1.2.2 to match. release_plan.py --check and bootstrap_audit.py --check pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Carry #303's dev sync (core 1.2.2, MCP 0.1.3, Vertex AI 0.1.11). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> # Conflicts: # python/pyproject.toml
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Completes the metadata used by PyPI, npm, and NuGet on each package's next publish, and fixes confirmed npm package entry points in a separate conventional commit. Stacked on #303 through a normal branch merge, with explicit 0.1.0 → 0.1.1 bumps for exactly the 18 repaired npm packages. No package names, dependencies, or build scripts change. No PR was merged and nothing was published.
fi-instrumentation-oteland its version unchanged.npm packaging repair and explicit 0.1.1 releases
Commit
b4414b1is the separatefix(npm)repair. All 47 packages were built and actually packed. Eighteen tarballs declared targets they did not contain; only those eighteen entry maps changed.Separate commit
4c55284(fix(npm): release the entry-point repairs as 0.1.1) synchronizes each package.json, matching release-please manifest entry, runtime version literal, and new CHANGELOG entry. Live read-only npm checks confirmed 0.1.1 was unpublished for every package before editing. The original 38 bootstrap patches remain intact; an additional 18 explicit-bump floor records preserve this repair without rewriting the historical audit snapshot.Why explicit bumps: on the first main run, the version-change publisher runs before release-please. Leaving these at the already-published 0.1.0 would skip the upload, reconcile a 0.1.0 GitHub release at the repair commit, and cause release-please to treat the fixes as already released. The end-to-end simulation reproduced that failure; explicit 0.1.1 versions make all 18 real upload candidates instead.
The following all now declare 0.1.1:
@traceai/fi-instrumentation-google-adk@traceai/fi-instrumentation-livekit@traceai/fi-instrumentation-pipecat@traceai/fi-instrumentation-vertexai@traceai/beeai,@traceai/chromadb,@traceai/guardrails,@traceai/instructor@traceai/lancedb,@traceai/milvus,@traceai/mongodb,@traceai/pgvector@traceai/pinecone,@traceai/portkey,@traceai/qdrant,@traceai/redis@traceai/strands,@traceai/weaviateThe four
fi-instrumentation-*packages now point to the emitteddist/srcfiles. For all eighteen, bothrequireand nativeimportresolve to the same CommonJS implementation, with named exports verified. Invalidmodule/esnextdeclarations are removed. These packages are CommonJS-only until their ESM builds emit Node-valid specifiers; changing/indexto/src/indexalone would still leave extensionless ESM imports broken. The other 29 packages' entry maps are unchanged. Published tarballs are immutable: this PR cannot repair an existing0.1.0installation until a new version is released.Provider runtime limits are separate from packaging
The new pages no longer present registration as proof that the SDK is instrumented:
manuallyInstrument()do not capture those calls.python3 -m pip install guardrails-ai==0.4.2. Documents Python/pip and an isolated writable virtual environment. Python installation and live validation were not executed.These four pages provide an explicit ESM application-span fallback, verified with mocked asynchronous SDK calls and actual export to a loopback OTLP receiver. It records duration/status, not automatic prompts, tools, token usage, or cost. Streaming must be consumed inside the callback. Provider runtime implementation fixes are not included in this metadata PR.
Verification
Round 3: stacked release and explicit npm patches
release_plan.py --check,bootstrap_audit.py --check, and the real release-please 17.6.0 strategy test pass.npm_release.publish()function with subprocess writes mocked select 18 uploads of real packed 0.1.1 tarballs and 18 matching GitHub releases. No upload or release creation was performed.release-pr --local --dry-runagainst simulated main, with only the LocalGitHub creation hook and in-memory publisher releases, leaves all 18 at 0.1.1, rewriting none of their package files. A local-only synthetic subsequent fix advances all 18 to 0.1.2. Network mutations and local tag creation are blocked; the synthetic commit is not pushed..ctsand.mts) and bundler. 133 checks pass, with no missing declared targets.Earlier metadata and runtime verification
moduleResolution: node16in both.ctsand.mts, and undermoduleResolution: bundler.agent.run,client.chat.completions.create,guard.parse, andagent.invokeoperations. No live provider requests or Cloud export were used.git diff --check: pass.Registry comparison
Compared live PyPI metadata for OpenInference OpenAI, Langfuse, and Braintrust, and npm metadata for OpenInference OpenAI, Langfuse tracing, and Braintrust. Used their populated discovery fields as the benchmark: meaningful summaries/Markdown READMEs, project/documentation/source/issue links, keywords, license, author/company, and repository directories. Registry pages update on release, not merely when this PR merges.
Existing blockers outside these fixes
python/traceaideclarestraceai, but the live registry project is Polyaxon's unrelated0.0.0package. Do not publish this helper under that name. Its README uses a verified source installation; no rename was made.python/traceaiusespackages.from = "..". Rebuilding its wheel from its sdist fails on unchanged dev too; direct wheel building and source installation work. Package restructuring is outside this lane.Merge order and overlap with #303
PR targets dev and now includes #303 at
896d14eb530ea38d449f8a4b14b523530b17d2e2via normal merge commit5a726c2. Merge #303 first; after that #306's diff shrinks to its own metadata, packaging repairs, and explicit npm patches. Do not merge the release automation twice or squash away its ancestry before updating this branch.The two known conflicts are already resolved on this branch, keeping #303's release versions and this PR's metadata:
python/frameworks/llama_index/pyproject.toml: version 0.1.10, description LlamaIndex.python/frameworks/smolagents/pyproject.toml: version 0.1.6, description smolagents.All 115 Python/npm metadata manifests retain their #306 non-version fields. All 38 original bootstrap patch versions remain identical to #303. No PR has been merged.
Evidence
Reproducible scripts, actual tarballs, wheel metadata, local OTLP captures, and per-package logs are under
/Volumes/nik-ext-1/tmp/traceai-sdk-release-20261029/lanes/metadata/evidence/. Round 3 evidence is inr3/:npm-0.1.1-availability.json,bump-scope.json,merge-preservation.json,publisher-simulation.json,publisher-order-result.json,cli-after-publishers.log,cli-follow-up.log,strategy-17.6.0.log, andpacked/results.json. Earlier evidence remains infix-npm-builds,fix-final-packs,fix-runtime,fix-python, andfix-static.json. Evidence is not shipped in the packages. No registry login/publication, new tags, GitHub releases, settings/secrets changes, force-push, or PR merges were performed.🤖 Generated with Claude Code