Release v2.6.0: Answer API, Speakeasy removal, flattened API surface - #31
Closed
tyler5673 wants to merge 20 commits into
Closed
Release v2.6.0: Answer API, Speakeasy removal, flattened API surface#31tyler5673 wants to merge 20 commits into
tyler5673 wants to merge 20 commits into
Conversation
tyler5673
force-pushed
the
release/v2.6.0
branch
5 times, most recently
from
August 5, 2026 21:43
39b1c45 to
5eb0872
Compare
![factory-droid[bot]](https://avatars.githubusercontent.com/in/358017?s=60&v=4){kind=small}
### Added - Answer API: you.answer() / you.answer_async() for POST /v1/answer. Returns synthesized markdown answers with inline citations and web results. - PaymentRequiredResponseError: first-class 402 error class for the answer API (UpgradeRequiredResponse schema). ### Changed - Flattened API surface: sub-SDK chains (you.search.unified(), you.contents.generate()) are deprecated with DeprecationWarning and delegate to new direct methods (you.search(), you.contents()). Plain strings accepted instead of enum imports for country, language, safesearch, livecrawl, freshness. - Removed all Speakeasy generated code and disclaimers. SDK is now fully hand-maintained. - Removed YDCUserAgentOverrideHook (no-op after Speakeasy removal). __user_agent__ derived from __version__ at runtime. - search_helpers module merged into you.search() direct method. - 422/500 error data models expanded with optional detail/errors fields (backward compatible). - Dev dependencies updated: mypy >=2.3.0, pylint >=4.0.0, pytest >=9.0.0, pytest-asyncio >=1.0.0. - CI: replaced pylint/pyright with mypy, excluded live tests from CI. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
tyler5673
force-pushed
the
release/v2.6.0
branch
from
August 5, 2026 22:20
5eb0872 to
75115cd
Compare
tyler5673
requested review from
EdwardIrby,
gvelez17,
itsakhilyou and
richarddavid-you
August 5, 2026 22:25
… empty api_key_auth opt-out - Fix broken MIGRATION.md anchor link (#250-to-260 → #250--260) - CONTRIBUTING.md: add --ignore=tests/test_performance.py to unit test command - Security: explicit empty api_key_auth now constructs Security(api_key_auth=None) instead of falling back to env vars (prevents confused-deputy footgun) Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Changes look cohesive and prior review feedback appears addressed; I did not find additional high-confidence, inline-anchorable issues beyond existing PR comments. |
…t_live.py - Delete .agents/skills/generate-sdk-and-open-pr/SKILL.md — was a Speakeasy generation workflow, no longer relevant since SDK is hand-maintained - Remove empty .agents/ directory tree - Fix stale 'overlay-generated' comment in test_live.py Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid encountered an error —— View job Droid is reviewing code and running a security check… |
Keyless search never shipped in a release, so referencing a revert is misleading. Renamed heading and header to describe the current state without implying users experienced keyless. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid encountered an error —— View job Droid is reviewing code and running a security check… |
- Remove 'Hand-maintained additions' section from CHANGELOG 2.4.0 (references deleted registration.py, removed YDCUserAgentOverrideHook, and Speakeasy regen workflow that no longer exists) - Remove .speakeasy/ entries from .gitignore - Fix stale 'overlay' references in examples/api-example-calls.py, MIGRATION.md, tests/test_research.py - Fix stale 'reverted overlay' comment in tests/test_research.py - Fix stale 'Speakeasy regen' comment in tests/test_security_env.py - Clean up Speakeasy reference in CHANGELOG 2.4.0 Fixed section Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
The POST /v1/search endpoint was already supported in 2.5.x. The actual change is search_helpers.search() became you.search(). Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Validated 4 inline comments for this PR: one correctness issue in |
- Hit /_mockserver/health (returns 200) with retry loop instead of broken 'curl -sf / || echo ready' that always printed ready - Remove -x flag so CI shows all failures, not just the first one Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid encountered an error —— View job Droid is reviewing code and running a security check… |
- scripts/check_drift.py: fetches OpenAPI specs from you.com and compares endpoints, server URLs, and enum values against the SDK surface - .github/workflows/test.yml: non-blocking drift check on every PR - .github/workflows/drift-check.yml: weekly scheduled check that opens a GitHub issue if drift is detected, closes stale issues when clean Known uncovered APIs (billing, images) and endpoints (GET /v1/search) are excluded via exception lists. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid encountered an error —— View job Droid is reviewing code and running a security check… |
- New handler: tests/mockserver/internal/handler/pathpostv1answer.go - Register route in generated_handlers.go - Add TestAnswerMockServer class with 4 tests hitting the mock server (basic, freshness, boost_domains, async) - Existing MockTransport tests retained for error case coverage - Update tests/README.md test count (19 -> 23) Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Phase 2 validation complete.
Artifacts: |
- drift-check.yml: add explicit permissions (contents: read, issues: write) so GITHUB_TOKEN can create/close issues in scheduled runs - check_drift.py: use context manager for httpx.Client to guarantee cleanup on exception paths Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Phase 2 (validator) complete
Large, cohesive v2.6.0 release refactor with substantial SDK, CI, and docs changes. No additional high-confidence, inline-anchorable correctness or security issues found beyond existing PR discussions. |
Drift check (check_drift.py): - Add request parameter drift: compares spec request body properties against SDK method signatures - Add response schema drift: compares spec 200 response schema fields against SDK pydantic model fields - Fix resolution for request bodies - Fix array response schema handling (contents returns List[]) Real drift found and fixed: - ResearchResponse missing 'warnings' field (List[str]) per OpenAPI spec - Added to model and TypedDict, backward-compatible (Optional, defaults None) CI improvements (test.yml): - Add Python 3.11 and 3.13 to test matrix (was 3.10 + 3.12 only) - Add pytest-cov coverage reporting (--cov=youdotcom --cov-report=term-missing) - Add pylint errors-only check (non-blocking, continue-on-error) - Add package build check job (python -m build --sdist --wheel) Dependabot (.github/dependabot.yml): - Weekly checks for pip and github-actions dependency updates Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Phase 2 (validator) complete.
Validated 6/6 inline candidates for this PR; the main actionable items are HTTP client lifecycle cleanup (especially for async context-manager usage and tests) plus a couple of CI/docs accuracy fixes to avoid misleading outputs. |
… drift workflow P1: You.__aexit__ now closes SDK-owned sync client too; __exit__ closes SDK-owned async client (prevents socket leaks in async-only usage) P2: Fix Security.serialize_model() optional_fields — was 'ApiKeyAuth' but actual field key is 'api_key_auth' (no alias), causing null fields to be re-introduced despite exclude_none P2: Drift workflow now distinguishes 'DRIFT DETECTED' from 'ERROR:' before opening an issue (prevents false drift issues on network failures) P2: Search docs — removed stale GET recommendation and changed UsageSnippet method from get to post P2: Document test client cleanup expectations in create_test_http_client Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Cohesive v2.6.0 refactor with a clear API split, but two correctness issues remain: sync context-manager teardown can leak the SDK-owned AsyncClient under a running event loop, and the documented global (Posted 2 inline review comments.) |
P1: Fix __exit__ async client close under running event loop — use get_running_loop() + run_coroutine_threadsafe() pattern (matching close_clients()) instead of asyncio.run() which raises RuntimeError in notebooks/async contexts P1: Clarify server_url constructor docstring — search/contents default to ydc-index.io and are not affected by the constructor server_url unless the per-method server_url argument is passed Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Phase 2 (validator) results
Overall this release PR looks cohesive, and the only remaining issue I found is a docs/type mismatch for Finance Research effort levels. |
- CHANGELOG: Remove 'Search/Contents host' from Changed (was already the case in 2.5.0, not a new change in 2.6.0) - CHANGELOG: Reword 'search() server_url fix' to accurately state that constructor server_url does NOT affect search/contents (only per-method server_url is respected) - README: Fix error count from 27 to 28 (3 httpx + 25 YouError subclasses) - MIGRATION: Add note that server URL behavior was already the case in 2.5.0, documented for reference - tests/mockserver/README: Replace stale 'auto-generated code from Speakeasy' with 'hand-maintained', remove deleted pathgetv1search.go from handler list, add all current handler files - PR description: Remove 'Search/Contents host split' from Breaking Changes (was already the case in 2.5.0) Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
FinanceResearchEffort no longer includes a LITE tier (only DEEP and EXHAUSTIVE per official You.com docs). Removed stale 'lite' bullet from docstrings in finance_researchop.py (2 places) and sdk.py finance_research/finance_research_async (2 places). Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Phase 2 (validator) complete
Large, cohesive v2.6.0 release refactor across SDK, CI, and docs. No additional high-confidence, inline-anchorable correctness or security issues found beyond existing PR discussions. Artifacts: |
P1: Fix drift-check.yml re-execution TOCTOU race — capture first run's
output once via GITHUB_ENV instead of re-running the script a second
time. Use distinct exit codes (0=no drift, 1=drift, 2=fetch error)
instead of grepping free-form stdout, eliminating the silent-loss
fallback where neither DRIFT DETECTED nor ERROR: appeared.
P1: Fix test_answer.py resource leaks — test_async_answer was creating
a real httpx.AsyncClient() (no MockTransport) that opened TCP sockets
to the mock server and was never closed. Switched to MockTransport
pattern (matching the rest of the file) with explicit try/finally
cleanup. Also added client.close() to the 3 sync mock-server tests
that were leaking create_test_http_client instances.
P2: Fix check_drift.py request param schema extraction — was using
bare schema.get('properties', {}).keys() while the response path used
_get_schema_properties() which handles oneOf/anyOf/allOf and arrays.
Unified to use _get_schema_properties for both paths. Also removed
redundant inline 'import re' in fetch_specs() (already at module level).
P2: Add upper bounds to pyproject.toml dev deps — mypy, pylint, pytest,
pytest-asyncio were unbounded across semver-major bumps. Pinned upper
bounds (<3, <5, <10, <2) so Dependabot can't float untested majors.
P3: Add failure sentinel to test.yml mock-server readiness loop — if
all 10 retries fail, the loop now exits 1 with a clear error message
instead of silently falling through to pytest.
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Contributor
|
Droid finished @tyler5673's task —— View job Phase 2 (validator) complete
Phase 2 validation complete. No inline candidates to validate or post; no additional high-confidence, inline-anchorable issues found beyond existing PR discussion. |
Addresses the review findings on the v2.6.0 release PR.
Versioning:
- Bump 2.6.0 -> 3.0.0. The Agents API, `search_post()`, `search_helpers`,
and the sub-SDK classes are removed, which is breaking and needs a major
version under the SemVer policy the CHANGELOG claims to follow.
Behavior:
- An empty `api_key_auth` now raises ValueError at construction instead of
silently falling back to the environment. Every endpoint requires a key,
so `""` is never valid -- it means a key was expected and none arrived,
nearly always `os.getenv("YDC_API_KEY", "")` with the variable unset.
Falling back there ran the request under a different identity than the
code asked for. Callables are checked on first use.
- `search(language=None)` sends no language again. Omitting the argument
still uses the API default (EN); previously an explicit None also
resolved to EN, with no way to opt out of the filter.
- `__exit__`/`__aexit__` no longer let a failing client close replace the
exception propagating out of the block, matching `close_clients`.
- `country`/`language`/`safesearch`/`livecrawl`/`livecrawl_formats`/
`freshness` are now genuinely case-normalized, which the docs already
claimed. Also fixes the `to` separator in uppercase date ranges.
- `search_async()`/`contents_async()` are real methods with full
signatures instead of `**kwargs: Any` wrappers that erased them for type
checkers and IDEs. Clears the one pylint error in the tree.
Tests (154 -> 249):
- New coverage for debug-log redaction, empty-key rejection, dual client
close and its non-masking property, and the normalization matrix
including the deprecated `unified()` path. Redaction and the key
handling were both shipped untested.
Docs:
- Replace all 52 `os.getenv("YDC_API_KEY", "")` examples, which the new
key handling turns into an error.
- CHANGELOG gains Removed/Security/Fixed sections; drops an entry for a
file that never existed on main and two no-op entries. Security is now
redaction alone -- the key change is fail-fast, not a closed hole.
- MIGRATION gains an action-required table plus sections on key
resolution, client lifecycle, redaction, and `language`.
CI:
- Drift issues update an existing open issue instead of filing a duplicate
weekly; body written via --body-file so the code fence renders; a broken
checker exits 3 and fails the job rather than reporting "no drift"
forever; --strict and --verbose mean something again.
- Dev tooling installs from the pyproject dependency group instead of a
duplicated list. pylint gates now that the tree is clean. dependabot
switched pip -> uv so uv.lock actually gets updated.
Verified: 249 tests pass, mypy clean (80 files), pylint 10.00/10, drift
check green against the live specs.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Contributor
|
Droid finished @tyler5673's task —— View job Phase 2 (validator) complete
Large, cohesive 3.0.0 release refactor (Answer API, Speakeasy removal, CI/drift tooling, and agent removal). No additional high-confidence, inline-anchorable correctness or security issues found beyond existing PR discussions. |
Contributor
Author
|
Superseded by #32. Same work, reopened on |
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
1 participant
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.
Summary
Major release with four themes: new Answer API, complete Speakeasy removal, flattened API surface, and drift detection.
New Features
you.answer()/you.answer_async()forPOST /v1/answeronapi.you.com. Returns synthesized markdown answers with inline citations, a citations array (source URLs + excerpts), and web results. Acceptsquery,freshness,country,language,include_domains,exclude_domains,boost_domains. Country/language accept plain strings (normalized to uppercase).PaymentRequiredResponseError: New first-class error class for HTTP 402 responses with structured data (error,message,upgrade_url,limit,used,period,reset_at).UnprocessableEntityResponseErrorDatanow includes optionaldetail(FastAPI) anderrors(JSON:API) fields.InternalServerErrorResponseDataadds optionalerrorsfield. Backward compatible.scripts/check_drift.py): Fetches You.com OpenAPI specs and compares endpoints, server URLs, and enum values against the SDK surface. Runs as a non-blocking CI step on every PR and as a weekly scheduled workflow that opens an issue if drift is detected.Breaking Changes
you.agents(),you.agents_async(),you.agents.runssub-SDK shim, and all agent model/error classes deleted. No agent-related code remains insrc/ordocs/.search_helpersmodule removed: Useyou.search(query=...)directly.you.search_post()alias removed: Useyou.search().you.search.unified()andyou.contents.generate()still work via backward-compat shims but emitDeprecationWarning. Preferred:you.search()andyou.contents().Speakeasy Removal
__gen_version__/SPEAKEASY_GENERATOR_VERSIONexports removed.YDCUserAgentOverrideHookdeleted;__user_agent__derived from__version__._hooks/registration.py(no-opinit_hooks) deleted.overlays/python_overlay.yamldeleted..speakeasy/entries removed from.gitignore.Security
_redact_headers()inbasesdk.pyredactsAuthorization,X-API-Key,Cookie,Set-Cookiebefore logging. Applied to both sync and async paths.api_key_authopt-out: Explicit empty string constructsSecurity(api_key_auth=None)instead of falling back to env vars, preventing confused-deputy behavior.Bug Fixes
SDKConfiguration.retry_config: Fixedpydantic.Fieldtodataclasses.field(stdlib@dataclassdoes not interpretFieldInfo)._populate_from_globals: Fixedis notto!=for string comparison (identity check could silently fail depending on interning).Dead Code Cleanup
.py+ 3 error.py).models/__init__.pyanderrors/__init__.pyexports cleaned up._shims.py: hoisted duplicated_split_csvto module level..agents/skills/generate-sdk-and-open-pr/SKILL.mddeleted (stale Speakeasy generation workflow).CI Improvements
/_mockserver/healthwith retry loop instead of brokencurl -sf /).-xflag so CI shows all failures, not just the first.Test Coverage