Repository navigation
fix(claude-agents)!: honour the config's max_tokens and disable thinking unless the config asks - #237
Conversation
…ing unless the config asks
For AIC-3585. The handler ran with the Claude Code CLI's own defaults: the
config's max_tokens was dropped, and with no thinking setting the CLI chose
one (a 31,999-token budget on claude-haiku-4-5).
max_tokens now reaches the CLI as CLAUDE_CODE_MAX_OUTPUT_TOKENS in env,
which the Agent SDK adds to os.environ. When the config sets neither
thinking nor max_thinking_tokens, the handler passes
thinking={"type": "disabled"}.
BREAKING CHANGE: a config that sets no thinking now runs without thinking.
Configs that relied on the CLI's adaptive-thinking default must set
thinking in their model parameters, for example {"type": "adaptive"}.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…'s test gaps Drop a max_tokens above 2^53 - 1, as the JS SDK does for a value that is not a safe integer, so both SDKs send the same thing for the same config. Tests: the environment the Agent SDK builds for the CLI still holds os.environ; effort alone and thinking with max_thinking_tokens; and a native-graph node's CLI command carries --thinking disabled, plus the invalid max_tokens and max_thinking_tokens-only node cases. README notes that a settings-file env wins and that some models reject disabled thinking. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jeffdupont
left a comment
There was a problem hiding this comment.
This matches mono#110 and js-ai-sdk#183. The same max_tokens bounds (a positive whole number up to 2^53 - 1, with 200.0 read as 200), the same disabled-thinking default, and the same rule that a dropped malformed thinking counts as not set. I checked env against the SDK: no host or skills option also sets env, so nothing overwrites it. One small test gap, below, is the same one I raised on the JS PR.
…ormed thinking and effort alone
Match the JS graph tests: a node with a thinking that has no budget, and
a node with only effort, each get thinking={"type": "disabled"}.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jeffdupont
left a comment
There was a problem hiding this comment.
cd91993 covers the test gap from my last review: the native-graph test now pins the disabled default for a malformed thinking ({"type": "enabled"} with no budget) and for effort alone, so every output-limit case mono#110 lists runs through invoke, stream and a graph node. Nothing else changed since 837053d, and the behaviour still matches js-ai-sdk#183.
Steps 1 and 2 of [AIC-3672](https://launchdarkly.atlassian.net/browse/AIC-3672). The JS half is AIC-3673. ## What this does - **`release-as: 1.0.0`** on all 14 packages in `release-please-config.json`. Without it, release-please keeps bumping 0.x minors, so the next release PR would ship `launchdarkly-ai-server` 0.6.0 instead of 1.0.0. - **Internal pins.** Every package that depends on `launchdarkly-ai-server` now requires `>=1.0.0,<2`. release-please doesn't update these for Python. #218 (merged) set them to `>=0.6,<2` so either release order resolves; this tightens them to the 1.0 line. - **`uv.lock`** needs no change: uv records workspace dependencies without a version specifier. `uv lock --check` passes. `changelog-sections` is left alone. #235 removes the evaluations Experimental rows when it promotes Evaluations from Code to core. ## Merge this last Merge this after the PRs below. Anything merged before the release PR (#231) merges ships in 1.0.0, and a `!` change merged after 1.0 forces 2.0. - [x] #218 fix(AIC-3598)!: packaging and public types (merged 29 Oct) - [x] #235 feat(evaluations)!: promote Evaluations from Code to core (merged 29 Oct). After it merges, nothing in `launchdarkly_ai_server.experimental` is public. - [x] #237 fix(claude-agents)!: honour max_tokens and disable thinking by default (merged 29 Oct) - [ ] #136 feat(AIC-3439): wildcard handlers accept only named providers (approved; conflicts with main, needs a merge of main) - [x] #234 fix(AIC-3568)!: give the barrel core's `__all__` (merged 29 Oct) - [ ] #246 fix(AIC-3592): make native LangGraph hand-offs complete (spec launchdarkly/ai-sdks-monorepo#118) - [ ] #248 fix(AIC-3573): never send a NativeTool to the model as a function tool (spec launchdarkly/ai-sdks-monorepo#116) - [ ] #240 feat(client): add result variant and value to the feature_flag span event (approved, mergeable) - [ ] #251 fix(AIC-3716): build OpenAI native graph node tools with FunctionTool (review required; touches the same loop as #248) #241 (Shadow Requests) is closed; 1.0 ships with nothing experimental. ## After merging 1. Confirm the regenerated #231 lists 1.0.0 for all 14 packages, then merge it. 2. Install each package from PyPI into an empty venv and import its public entry points (AIC-3672 step 5). 3. Open a PR removing `release-as` straight away, or every later release PR proposes 1.0.0 again. ## Checks - `release-please-config.json` parses as JSON, with `release-as: 1.0.0` on all 14 packages. The first push wrote the lines with literal backslashes; fixed in d720896. - `make test`: 3,711 passed, 18 skipped (before #175 and #236 merged; CI is the current check) (with the parent checkout's `.env` LD URIs blanked; see AIC-3593) - `uv lock --check` passes 🤖 Generated with [Claude Code](https://claude.com/claude-code) [AIC-3672]: https://launchdarkly.atlassian.net/browse/AIC-3672?atlOrigin=eyJpIjoiNWRkNTljNzYxNjVmNDY3MDlhMDU5Y2ZhYzA5YTRkZjUiLCJwIjoiZ2l0aHViLWNvbS1KU1cifQ <!-- CURSOR_SUMMARY --> --- > [!NOTE] > **Overview** > Prepares the **1.0.0 GA** cut for all 14 Python packages by teaching release-please to release **1.0.0** instead of another **0.x** bump. > > Adds **`release-as: "1.0.0"`** on every entry in `release-please-config.json` (core `launchdarkly-ai-server`, barrel, and all handler packages). > > Tightens **internal PyPI pins** from `launchdarkly-ai-server>=0.6,<2` to **`>=1.0.0,<2`** in each handler `pyproject.toml` and in `packages/ai` (including **`otel`** / **`ld`** extras). Published wheels will require the 1.0 server line once those versions ship. > > <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit ae9f2cc. Bugbot is set up for automated code reviews on this repo. Configure [here](https://www.cursor.com/dashboard/bugbot).</sup> <!-- /CURSOR_SUMMARY --> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: andrewklatzke <aklatzke@launchdarkly.com>
🤖 I have created a release *beep* *boop* --- <details><summary>launchdarkly-ai-server: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-server-0.5.0...launchdarkly-ai-server-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **client:** init_client raises when the LaunchDarkly client did not initialize ([#254](#254)) * **evaluations:** route generation and judges to a scoped wildcard handler only for its named providers, and accept two scoped wildcards of one mode when their lists differ * **AIC-3439:** let wildcard handlers accept only named providers * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **evaluations:** promote Evaluations from Code to core ([#235](#235)) * **client:** capture conversation content by default ([#213](#213)) * **evaluations:** remove default pass rate threshold ([#238](#238)) * **evaluations:** set default pass rate threshold ([#226](#226)) * **client:** spans carry launchdarkly.project_id instead of highlight.project_id * **AIC-3687:** rename create_openai_agent_handler to create_openai_agents_handler ([#225](#225)) * **skills:** rename FDv2SkillStore to LaunchDarklySkillStore ([#223](#223)) ### Features * **AIC-3439:** let wildcard handlers accept only named providers ([454a4de](454a4de)) * **AIC-3687:** rename create_openai_agent_handler to create_openai_agents_handler ([#225](#225)) ([a7cdbc0](a7cdbc0)) * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) * **client:** point judge spans at the run they graded ([#253](#253)) ([863d423](863d423)) * **evaluations:** promote Evaluations from Code to core ([#235](#235)) ([0bc43a5](0bc43a5)) * **evaluations:** remove default pass rate threshold ([#238](#238)) ([882d099](882d099)) * **evaluations:** route generation and judges to a scoped wildcard handler only for its named providers, and accept two scoped wildcards of one mode when their lists differ ([454a4de](454a4de)) * **evaluations:** set default pass rate threshold ([#226](#226)) ([7ee368c](7ee368c)) * **skills:** rename FDv2SkillStore to LaunchDarklySkillStore ([#223](#223)) ([11f2c0b](11f2c0b)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) * **AIC-3601:** stop graph routing text inviting a router to answer for its target ([#230](#230)) ([27b8d09](27b8d09)) * **client:** init_client raises when the LaunchDarkly client did not initialize ([#254](#254)) ([6e66c00](6e66c00)) * **client:** stamp launchdarkly.project_id on the telemetry resource ([#229](#229)) ([5347558](5347558)) * **evaluations:** copy inline tool schemas and name both sources on a key collision ([#227](#227)) ([bfbdd4c](bfbdd4c)) * **evaluations:** send criterion errors as errorMessage and skip criteria for rows that did not generate ([#228](#228)) ([8b05b56](8b05b56)) * **graph:** report a node's own judge scores on its node_done stream event ([#236](#236)) ([2f3f9ee](2f3f9ee)) ### Documentation * **AIC-3692:** say which project id attribute the telemetry resource carries ([#239](#239)) ([64eff43](64eff43)) * state that Agent Skills and Evaluations from Code need 1.0.0 ([#191](#191)) ([fe3ef08](fe3ef08)) </details> <details><summary>launchdarkly-ai-python: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-python-0.4.0...launchdarkly-ai-python-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **evaluations:** promote Evaluations from Code to core ([#235](#235)) ### Features * **evaluations:** promote Evaluations from Code to core ([#235](#235)) ([0bc43a5](0bc43a5)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) ### Documentation * state that Agent Skills and Evaluations from Code need 1.0.0 ([#191](#191)) ([fe3ef08](fe3ef08)) </details> <details><summary>launchdarkly-ai-claude-agents: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-claude-agents-0.3.0...launchdarkly-ai-claude-agents-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **claude-agents:** honour the config's max_tokens and disable thinking unless the config asks ([#237](#237)) * **client:** capture conversation content by default ([#213](#213)) * **AIC-3685:** make the internal provider helpers private ([#224](#224)) * **skills:** rename FDv2SkillStore to LaunchDarklySkillStore ([#223](#223)) ### Features * **AIC-3685:** make the internal provider helpers private ([#224](#224)) ([9afdb94](9afdb94)) * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) * **skills:** rename FDv2SkillStore to LaunchDarklySkillStore ([#223](#223)) ([11f2c0b](11f2c0b)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) * **claude-agents:** honour the config's max_tokens and disable thinking unless the config asks ([#237](#237)) ([a7ded1b](a7ded1b)) </details> <details><summary>launchdarkly-ai-claude-messages: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-claude-messages-0.3.0...launchdarkly-ai-claude-messages-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) * **AIC-3735:** send Claude Messages sampling parameters through extra_body ([#256](#256)) ([c6d0376](c6d0376)) </details> <details><summary>launchdarkly-ai-openai-agents: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-openai-agents-0.3.0...launchdarkly-ai-openai-agents-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) * **AIC-3685:** make the internal provider helpers private ([#224](#224)) * **AIC-3687:** rename create_openai_agent_handler to create_openai_agents_handler ([#225](#225)) ### Features * **AIC-3685:** make the internal provider helpers private ([#224](#224)) ([9afdb94](9afdb94)) * **AIC-3687:** rename create_openai_agent_handler to create_openai_agents_handler ([#225](#225)) ([a7cdbc0](a7cdbc0)) * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) * **AIC-3716:** build OpenAI native graph node tools with FunctionTool ([#251](#251)) ([b859b67](b859b67)) * **AIC-3726:** realign three usage IDs with the 1.0 factory names ([#252](#252)) ([1c4d904](1c4d904)) </details> <details><summary>launchdarkly-ai-vercel-agents: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-vercel-agents-0.2.0...launchdarkly-ai-vercel-agents-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) </details> <details><summary>launchdarkly-ai-litellm-agents: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-litellm-agents-0.2.0...launchdarkly-ai-litellm-agents-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) </details> <details><summary>launchdarkly-ai-openai-messages: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-openai-messages-0.3.0...launchdarkly-ai-openai-messages-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) * **AIC-3726:** realign three usage IDs with the 1.0 factory names ([#252](#252)) ([1c4d904](1c4d904)) </details> <details><summary>launchdarkly-ai-vercel-messages: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-vercel-messages-0.2.0...launchdarkly-ai-vercel-messages-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) </details> <details><summary>launchdarkly-ai-litellm-messages: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-litellm-messages-0.2.0...launchdarkly-ai-litellm-messages-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) </details> <details><summary>launchdarkly-ai-langchain-agents: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-langchain-agents-0.3.0...launchdarkly-ai-langchain-agents-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **evaluations:** route generation and judges to a scoped wildcard handler only for its named providers, and accept two scoped wildcards of one mode when their lists differ * **AIC-3439:** let wildcard handlers accept only named providers * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) ### Features * **AIC-3439:** let wildcard handlers accept only named providers ([454a4de](454a4de)) * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) * **evaluations:** route generation and judges to a scoped wildcard handler only for its named providers, and accept two scoped wildcards of one mode when their lists differ ([454a4de](454a4de)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3590:** build Python langchain-agents with langchain.agents.create_agent ([#175](#175)) ([8218626](8218626)) * **AIC-3592:** make native LangGraph hand-offs complete ([#246](#246)) ([bfb738f](bfb738f)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) </details> <details><summary>launchdarkly-ai-langchain-messages: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-langchain-messages-0.3.0...launchdarkly-ai-langchain-messages-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **evaluations:** route generation and judges to a scoped wildcard handler only for its named providers, and accept two scoped wildcards of one mode when their lists differ * **AIC-3439:** let wildcard handlers accept only named providers * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) ### Features * **AIC-3439:** let wildcard handlers accept only named providers ([454a4de](454a4de)) * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) * **evaluations:** route generation and judges to a scoped wildcard handler only for its named providers, and accept two scoped wildcards of one mode when their lists differ ([454a4de](454a4de)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3573:** never send a NativeTool to the model as a function tool ([#248](#248)) ([73caeae](73caeae)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) * **AIC-3726:** realign three usage IDs with the 1.0 factory names ([#252](#252)) ([1c4d904](1c4d904)) </details> <details><summary>launchdarkly-ai-opencode-agents: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-opencode-agents-0.3.0...launchdarkly-ai-opencode-agents-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) </details> <details><summary>launchdarkly-ai-typesafe: 1.0.0</summary> ## [1.0.0](launchdarkly-ai-typesafe-0.2.0...launchdarkly-ai-typesafe-1.0.0) (2026-10-30) ### ⚠ BREAKING CHANGES * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) * **client:** capture conversation content by default ([#213](#213)) ### Features * **client:** add result variant and value to the feature_flag span event ([#240](#240)) ([040d487](040d487)) * **client:** capture conversation content by default ([#213](#213)) ([77a7cf5](77a7cf5)) ### Bug Fixes * **AIC-3568:** give the barrel core's __all__, and drop private-spec references ([#234](#234)) ([88e57e6](88e57e6)) * **AIC-3598:** make Python packaging and public types GA-ready ([#218](#218)) ([5860b40](5860b40)) </details> --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please). <!-- CURSOR_SUMMARY --> --- > [!NOTE] > **Overview** > This is a **Release Please** cut that moves every monorepo package to **1.0.0** and records the GA release in changelogs. It updates `.release-please-manifest.json`, each package’s `pyproject.toml` / `__version__`, and prepends **1.0.0** notes to every `CHANGELOG.md`. > > Handler and barrel packages now require **`launchdarkly-ai-server>=1.0.0,<2`**, so installs must upgrade the core client in lockstep with adapters. > > There is **no new application logic** in this diff—only release metadata. The changelog text still matters for reviewers: it summarizes breaking work already on `main`, including **Evaluations from Code promoted to core**, **conversation content captured by default**, **`init_client` failing when LD init fails**, telemetry **`launchdarkly.project_id`**, renames like **`LaunchDarklySkillStore`** and **`create_openai_agents_handler`**, tightened **wildcard handler routing**, and **GA packaging / public `__all__`** cleanup. > > <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit 2e904f7. Bugbot is set up for automated code reviews on this repo. Configure [here](https://www.cursor.com/dashboard/bugbot).</sup> <!-- /CURSOR_SUMMARY --> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Summary
AIC-3585. Spec: launchdarkly/ai-sdks-monorepo#110. The same change for TypeScript is launchdarkly/js-ai-sdk#183; Go is not affected.
claude-agentsran with the Claude Code CLI's own defaults instead of the AI Config's. The config'smax_tokenswas dropped, becauseClaudeAgentOptionshas no output-token field. With no thinking setting, the CLI chose one: a 31,999-token budget onclaude-haiku-4-5. Python wasn't checked live; the code path is the same as the TypeScript one that was._options_parameters(handler.py) is shared by the handler'sinvokeandstreamand byto_claude_agents. It now:max_tokensto the CLI. A positive integermax_tokensup to 2^53 − 1 becomesenv={"CLAUDE_CODE_MAX_OUTPUT_TOKENS": "<n>"}. That's the same bound as the JS SDK's safe-integer check. The Agent SDK adds that to the inheritedos.environ. A whole-number float such as200.0counts as 200, as it does in JavaScript. Any other value is dropped. With nomax_tokens,envstays{}and the CLI keeps its default, as before.thinkingnormax_thinking_tokens, it passesthinking={"type": "disabled"}. A malformedthinkingthat gets dropped counts as not set. An explicitthinking, ormax_thinking_tokenson its own, is passed as before. The CLI leavesthinkingout for models that reject disabled thinking, such asclaude-fable-5.max_tokensis read separately rather than added to_CLAUDE_AGENT_OPTIONS_FORWARDED_KEYS, since it isn't aClaudeAgentOptionsfield and the drift test checks that list against the dataclass.envstays in the excluded list: themax_tokensvalue is the only thing written into it.tests/forwarding_spec.pyaddsmax_tokensto the Claude Agents list.Breaking change
A config that sets no thinking now runs without thinking. Before, the CLI chose: adaptive thinking on models that support it, or a fixed budget on others. Configs that relied on that must set
thinkingin their model parameters, for example{"type": "adaptive"}. The README's new "Output limit and thinking" section has the migration note.Tests
TestOutputLimitAndThinkingDefault, on bothinvokeandstream, covers:max_tokens: 200sets the variable and disables thinking, checked against the SDK's own CLI command builder.os.environ.open_processis stubbed to capture it.200.0is read as 200.max_tokens,envis empty.0,-5,1.5,2**53,inf,"200",True,Noneand a dict are each dropped.effortalone gets the default, andthinkingwithmax_thinking_tokenspasses both as given.thinkingreplaces the default.max_thinking_tokenson its own leavesthinkingunset.TestNativeGraphModelParameterscovers four node cases:max_tokens(and checks that the real SDK builds--thinking disabledfor it), its ownthinking, an invalidmax_tokens, andmax_thinking_tokensalone (--max-thinking-tokens 2048).TestThinkingIsRebuilt/TestMalformedObjectValuesAreDroppedcases expected a dropped thinking to beNone, and now expect the default.max_tokenswas removed fromtest_ui_keys_the_sdk_rejects_are_dropped_without_raising, since it's no longer dropped.make testpasses (3686 passed, 18 skipped).make lint,make format-checkandmypypass.🤖 Generated with Claude Code
Note
Overview
Breaking: configs that omit thinking no longer inherit Claude Code’s implicit thinking defaults; the handler now passes
thinking={"type": "disabled"}unless the config setsthinkingormax_thinking_tokens(malformedthinkingfalls back to the same default).The shared
_options_parameterspath (handler invoke/stream andto_claude_agents) now honoursmax_tokens: valid positive integers (including whole floats, capped at 2^53−1) are sent to the CLI asenv={"CLAUDE_CODE_MAX_OUTPUT_TOKENS": "<n>"}becauseClaudeAgentOptionshas no output-token field. Invalid values are ignored; omittingmax_tokensleaves CLI defaults unchanged.README documents output limits, thinking behaviour, and migration. Cross-SDK
forwarding_specaddsmax_tokensto the Claude Agents forwarded list; new handler, native-graph, and updated thinking-rebuild tests lock in the env/thinking rules and CLI command shape.Reviewed by Cursor Bugbot for commit cd91993. Bugbot is set up for automated code reviews on this repo. Configure here.