Release 0.4.0

[hiyouga]

Promote dev to main for the 0.4.0 release. Version is already bumped to 0.4.0 in both packages (#160).

Highlights since 0.3.3

• fidelity content-item payload (breaking) — replaces the per-item signature/phase fields with one opaque JSON payload of wire-fidelity data (Claude/Gemini signatures, GPT-5.5 {id, encrypted_content} and phase, OpenAI-compatible reasoning_field). OpenAI-compatible clients now replay thinking through exactly the reasoning field the upstream produced (reasoning_content vs reasoning) instead of always sending both spellings. Histories recorded by 0.3.x need signature/phase moved under fidelity to replay — see changelog/2026-07-20-reasoning-field-fidelity.md. (Add the fidelity payload and replay the exact reasoning field the upstream produced #159)
• Support Claude 5 models. (Support Claude 5 #149)
• Stronger streaming guarantees — the tool-call streaming protocol is documented and hardened (Document and harden the tool-call streaming protocol #152); malformed streamed tool arguments raise ToolCallArgumentParseError (fix: report malformed streamed tool arguments #155); thinking-only responses raise EmptyResponseError; both inherit the new AgentHubError base class (fix: reject thinking-only responses from reasoning models #157).
• Docs & tooling — agenthub-dev development-workflow skill and the changelog/ details directory; usage_metadata documented as a snapshot (docs(skills): usage_metadata is a snapshot, take the latest instead of accumulating #158); GPT-5.5 reasoning guide (including the phase parameter) snapshotted into llmsdk_docs/; package metadata links updated (Update package metadata links #150).

Both CI suites (Python 3.11 and Node 20, including live-API e2e) are green on dev as of #160.

Suggest merging with a merge commit (not squash) to preserve dev history, matching the 0.3.3 release, then tagging v0.4.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_01J8bNMvyPJ7muyDs4Fh9ZLF

[Copilot]

Pull request overview

Release 0.4.0 by promoting the dev line of changes to main, including a breaking data-model update to preserve wire-level replay fidelity across providers and tightening streaming/error behavior across both the Python and TypeScript SDKs.

Changes:

• Replace per-item signature / phase fields with a single opaque fidelity payload across content items (breaking; enables exact replay, including reasoning-field spelling).
• Harden streaming behavior: contextual tool-call JSON parse errors (ToolCallArgumentParseError) and rejection of thinking-only responses (EmptyResponseError) under a shared AgentHubError base class.
• Restructure skills documentation into a “reference/” layout, update changelog detail files, and bump package metadata to 0.4.0.

Reviewed changes

Copilot reviewed 58 out of 60 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src_ts/tests/tool-call-arguments.test.ts	Adds TS regression coverage for streamed tool-call argument parsing errors.
src_ts/tests/reasoning-fidelity.test.ts	Adds TS coverage for reasoning-field replay fidelity and phase-based concatenation.
src_ts/tests/empty-response.test.ts	Adds TS coverage for rejecting thinking-only / empty responses and error hierarchy.
src_ts/src/types.ts	Introduces Fidelity and migrates content items to fidelity?: ....
src_ts/src/openai/client.ts	Records/replays exact reasoning field; routes tool-call JSON parsing through shared helper.
src_ts/src/kimi_k2_6/client.ts	Same reasoning-field fidelity + tool-call parse hardening for Kimi.
src_ts/src/index.ts	Exports new error types from the TS package root.
src_ts/src/gpt5_5/client.ts	Moves phase/reasoning signature handling into fidelity; shared tool-call argument parsing.
src_ts/src/glm5_1/client.ts	Same reasoning-field fidelity + tool-call parse hardening for GLM.
src_ts/src/gemini3/client.ts	Migrates Gemini thought signatures into fidelity.signature.
src_ts/src/errors.ts	Adds AgentHubError, EmptyResponseError, ToolCallArgumentParseError, and parseToolCallArguments.
src_ts/src/deepseek_v4/client.ts	Records reasoning-field fidelity and uses shared tool-call argument parsing.
src_ts/src/claude5/index.ts	Exports Claude5Client.
src_ts/src/claude5/client.ts	Migrates Claude thinking signatures into fidelity.signature; shared tool-call argument parsing.
src_ts/src/claude4_6/client.ts	Migrates Claude thinking signatures into fidelity.signature; shared tool-call argument parsing.
src_ts/src/baseClient.ts	Updates concatenation rules to key off fidelity; validates non-thinking output.
src_ts/src/autoClient.ts	Updates TS auto-routing to include Claude 5 and avoids embedding misrouting.
src_ts/package.json	Bumps TS package to 0.4.0 and normalizes repository links.
src_ts/package-lock.json	Locks TS version bump to 0.4.0.
src_py/tests/test_tool_call_arguments.py	Adds Python regression coverage for streamed tool-call argument parsing errors.
src_py/tests/test_reasoning_fidelity.py	Adds Python coverage for reasoning-field replay fidelity and phase-based concatenation.
src_py/tests/test_empty_response.py	Adds Python coverage for rejecting thinking-only / empty responses and error hierarchy.
src_py/pyproject.toml	Bumps Python package to 0.4.0 and adds project URLs.
src_py/agenthub/types.py	Introduces Fidelity and migrates content items to fidelity.
src_py/agenthub/openai/client.py	Records/replays exact reasoning field; routes tool-call JSON parsing through shared helper.
src_py/agenthub/kimi_k2_6/client.py	Same reasoning-field fidelity + tool-call parse hardening for Kimi.
src_py/agenthub/gpt5_5/client.py	Moves phase/reasoning signature handling into fidelity; shared tool-call argument parsing.
src_py/agenthub/glm5_1/client.py	Same reasoning-field fidelity + tool-call parse hardening for GLM.
src_py/agenthub/gemini3/client.py	Migrates Gemini thought signatures into fidelity.signature and null-guards content parts.
src_py/agenthub/errors.py	Adds AgentHubError, EmptyResponseError, ToolCallArgumentParseError, and parse_tool_call_arguments.
src_py/agenthub/deepseek_v4/client.py	Records reasoning-field fidelity and uses shared tool-call argument parsing.
src_py/agenthub/claude5/client.py	Migrates Claude thinking signatures into fidelity.signature; shared tool-call argument parsing.
src_py/agenthub/claude5/init.py	Exports Claude5Client from the Python claude5 module.
src_py/agenthub/claude4_6/client.py	Migrates Claude thinking signatures into fidelity.signature; shared tool-call argument parsing.
src_py/agenthub/base_client.py	Updates concatenation rules to key off fidelity; validates non-thinking output.
src_py/agenthub/auto_client.py	Updates Python auto-routing to include Claude 5 and avoids embedding misrouting.
src_py/agenthub/init.py	Exports new Python error types from the package root.
skills/agenthub-typescript/SKILL.md	Rehomes model/docs content into reference pages; updates examples and fidelity guidance.
skills/agenthub-typescript/reference/models.md	Adds TS model-selection reference page.
skills/agenthub-typescript/reference/integrations.md	Adds TS tracer/playground integrations reference page.
skills/agenthub-typescript/reference/data-models.md	Adds TS data model + tool-streaming protocol + error reference page.
skills/agenthub-typescript/reference/api.md	Adds TS API reference page.
skills/agenthub-python/SKILL.md	Rehomes model/docs content into reference pages; updates examples and fidelity guidance.
skills/agenthub-python/reference/models.md	Adds Python model-selection reference page.
skills/agenthub-python/reference/integrations.md	Adds Python tracer/playground integrations reference page.
skills/agenthub-python/reference/data-models.md	Adds Python data model + tool-streaming protocol + error reference page.
skills/agenthub-python/reference/api.md	Adds Python API reference page.
README.md	Updates README examples to use fidelity.signature and notes Claude 5 in the matrix.
llmsdk_docs/gpt5_5/README.md	Adds reasoning doc link in GPT-5.5 snapshot index.
llmsdk_docs/gpt5_5/docs/reasoning.md	Snapshots reasoning/phase guidance for GPT-5.x reasoning models.
llmsdk_docs/claude5/README.md	Adds Claude 5 docs index.
llmsdk_docs/claude5/docs/introducing-claude-fable-5-and-claude-mythos-5.md	Snapshots Claude 5 release notes / behavior notes.
changelog/README.md	Introduces per-entry changelog detail file convention.
changelog/2026-07-20-reasoning-field-fidelity.md	Adds migration + rationale for the breaking fidelity change.
changelog/2026-07-17-agenthub-dev-skill.md	Documents addition of the agenthub-dev workflow skill and changelog details dir.
CHANGELOG.md	Adds 0.4.0 release line and links to detail files.
.gitignore	Ignores api_captures/ used for raw live API capture artifacts.
.agents/skills/agenthub-dev/SKILL.md	Adds a fixed workflow skill for model-support development in this repo.

Files not reviewed (1)

• src_ts/package-lock.json: Generated file

Comments suppressed due to low confidence (1)

src_py/agenthub/claude5/client.py:209

• item["fidelity"]["signature"] is accessed unconditionally for Claude thinking items. If a caller passes a history without fidelity.signature (e.g., hand-built history or pre-0.4.0 history that wasn’t migrated), this will raise a KeyError. Consider validating presence and raising a clearer error.