agentharnessapi 0.1.0__tar.gz → 0.1.1.dev2__tar.gz

agentharnessapi-0.1.1.dev2/.github/workflows/publish.yml

@@ -0,0 +1,139 @@
+# Publish to PyPI and GitHub Packages on every push/merge to main.
+#
+# Setup (one-time):
+#   1. Go to https://pypi.org -> Account Settings -> Publishing ->
+#      "Add a new pending publisher"
+#   2. Fill in:
+#        PyPI project name:  agentharnessapi
+#        Owner:              shibuiwilliam
+#        Repository:         AgenticAPI
+#        Workflow name:      publish.yml
+#        Environment name:   pypi
+#   3. Go to GitHub repo -> Settings -> Environments -> New environment
+#      named "pypi" (this gates the deployment with a manual approval
+#      if you add a reviewer, or runs automatically if you don't).
+#
+# PyPI uses "Trusted Publisher" (OIDC) -- no API tokens stored.
+# GitHub Packages publishes to a GitHub Release with the built
+# wheel and sdist attached, making the package installable via:
+#   pip install https://github.com/shibuiwilliam/AgenticAPI/releases/download/v0.1.0/agentharnessapi-0.1.0-py3-none-any.whl
+
+name: Publish to PyPI & GitHub Packages
+
+on:
+  push:
+    branches: [main]
+
+permissions:
+  contents: read
+
+jobs:
+  # -- Gate: full CI must pass before publishing ----------------------
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v6
+        with:
+          enable-cache: true
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.13"
+      - run: uv sync --group dev
+      - run: uv run ruff format --check src/ tests/ examples/
+      - run: uv run ruff check src/ tests/ examples/
+      - run: uv run mypy src/agenticapi/
+      - run: uv run pytest --ignore=tests/benchmarks --ignore=tests/e2e -q
+
+  # -- Build sdist + wheel --------------------------------------------
+  build:
+    needs: test
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0        # full history needed for hatch-vcs version
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.13"
+      - run: pip install build
+      - run: python -m build
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  # -- Publish to PyPI via Trusted Publisher (OIDC) -------------------
+  publish-pypi:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+      - uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          skip-existing: true
+
+  # -- Publish to GitHub Packages (GitHub Release) --------------------
+  #
+  # Creates a GitHub Release tagged with the package version and
+  # attaches the built wheel and sdist. This makes the package
+  # available via GitHub's package registry and direct URL install.
+  #
+  # Install from GitHub:
+  #   pip install https://github.com/shibuiwilliam/AgenticAPI/releases/latest/download/agentharnessapi-0.1.0-py3-none-any.whl
+  publish-github:
+    needs: build
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+      packages: write
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+      - name: Extract version from pyproject.toml
+        id: version
+        run: |
+          version=$(python3 -c "
+          import tomllib, pathlib
+          data = tomllib.loads(pathlib.Path('pyproject.toml').read_text())
+          print(data['project']['version'])
+          ")
+          echo "version=$version" >> "$GITHUB_OUTPUT"
+          echo "tag=v$version" >> "$GITHUB_OUTPUT"
+      - name: Create or update GitHub Release
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          tag="${{ steps.version.outputs.tag }}"
+          version="${{ steps.version.outputs.version }}"
+
+          # Check if the release already exists
+          if gh release view "$tag" > /dev/null 2>&1; then
+            echo "Release $tag already exists -- uploading artifacts"
+            gh release upload "$tag" dist/* --clobber
+          else
+            echo "Creating release $tag"
+            gh release create "$tag" dist/* \
+              --title "agentharnessapi $version" \
+              --notes "Published from commit ${{ github.sha }}.
+
+          Install from PyPI:
+          \`\`\`bash
+          pip install agentharnessapi==$version
+          \`\`\`
+
+          Install from GitHub:
+          \`\`\`bash
+          pip install https://github.com/${{ github.repository }}/releases/download/$tag/agentharnessapi-${version}-py3-none-any.whl
+          \`\`\`" \
+              --latest
+          fi

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/.gitignore

@@ -211,3 +211,6 @@ __marimo__/
 
 harness_playground_audit.sqlite
 agenticapi_memory_demo.sqlite
+
+# Auto-generated by hatch-vcs at build time
+src/agenticapi/_version.py

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/CLAUDE.md

@@ -8,7 +8,7 @@ AgenticAPI is a Python OSS framework that natively integrates coding agents into
 
 **Current status** (as of Increment 9, 2026-04-12). Core: **118 Python
 modules, ~21,944 LOC, 1,304 collected tests, 27 examples, 75 public
-exports**. Extensions: `agenticapi-claude-agent-sdk` v0.1.0 (~1,610 src
+exports**. Extensions: `agentharnessapi[claude-agent-sdk]` (~1,610 src
 LOC, 38 tests). **Phase A (control plane) is complete.** **Phase D
 (typed handlers + DI) core is complete** including schema-driven OpenAPI
 for typed `Intent[T]` request bodies (D7). Phases E / F have shipped
@@ -32,7 +32,7 @@ For the full plane-by-plane shipped / active / deferred matrix see
 ```bash
 uv sync --group dev              # Install all dependencies
 uv run agenticapi version        # Verify CLI works
-pip install -e ".[mcp]"          # Optional: MCP support
+pip install agentharnessapi[mcp]          # Optional: MCP support
 ```
 
 ### Testing
@@ -101,27 +101,24 @@ with their own `pyproject.toml`. Large or fast-moving dependencies
 stay out of the core package and live in extension packages so users
 only pay for what they use.
 
-**Current extensions:**
+**Current extras:**
 
-| Extension | Purpose | Deps |
-|---|---|---|
-| `agenticapi-claude-agent-sdk` | Wraps the [Claude Agent SDK](https://code.claude.com/docs/en/agent-sdk/overview) for full planning + tool-use loops inside agent endpoints | `claude-agent-sdk>=0.1.58,<0.2` |
+| Extra | Install | Purpose | Deps |
+|---|---|---|---|
+| `claude-agent-sdk` | `pip install agentharnessapi[claude-agent-sdk]` | Claude Agent SDK loop inside agent endpoints | `claude-agent-sdk>=0.1.58,<0.2` |
+| `mcp` | `pip install agentharnessapi[mcp]` | MCP server support | `mcp>=1.27.0` |
 
 ```bash
-# Install the extension for development (no-deps so main package stays linked)
-uv pip install -e extensions/agenticapi-claude-agent-sdk --no-deps
-
 # Run extension tests (offline — uses a stub SDK module via conftest.py)
-uv run pytest extensions/agenticapi-claude-agent-sdk/tests
+uv run pytest tests/unit/ext/claude_agent_sdk/
 
 # Type-check the extension
-uv run mypy extensions/agenticapi-claude-agent-sdk/src
+uv run mypy src/agenticapi/ext/
 ```
 
 See:
-- [docs/internals/extensions.md](docs/internals/extensions.md) — Extensions architecture and contribution guide
+- [docs/internals/extensions.md](docs/internals/extensions.md) — Extensions architecture
 - [docs/internals/claude-agent-sdk-extension-plan.md](docs/internals/claude-agent-sdk-extension-plan.md) — Claude Agent SDK extension design rationale
-- [extensions/agenticapi-claude-agent-sdk/README.md](extensions/agenticapi-claude-agent-sdk/README.md) — User-facing docs
 
 ---
 
@@ -221,7 +218,7 @@ See [docs/internals/modules.md](docs/internals/modules.md) for the complete modu
 | `PlainTextResult` | `interface/response.py` | Plain text response |
 | `UploadFile` | `interface/upload.py` | Uploaded file data (filename, content, size) |
 | `UploadedFiles` | `interface/upload.py` | Handler param type for auto-injected uploaded files |
-| `MCPCompat` | `interface/compat/mcp.py` | MCP server (`pip install agenticapi[mcp]`) |
+| `MCPCompat` | `interface/compat/mcp.py` | MCP server (`pip install agentharnessapi[mcp]`) |
 | `RESTCompat` | `interface/compat/rest.py` | REST route generation |
 | `HtmxHeaders` | `interface/htmx.py` | HTMX request header detection (injected into handlers) |
 | `htmx_response_headers()` | `interface/htmx.py` | Build HTMX response headers (HX-Trigger, etc.) |
@@ -439,7 +436,7 @@ See [examples/README.md](examples/README.md) for the full examples guide.
 | `10_file_handling` | None | File upload/download: `UploadedFiles`, `FileResult`, streaming |
 | `11_html_responses` | None | Custom responses: `HTMLResult`, `PlainTextResult`, `FileResult` |
 | `12_htmx` | None | HTMX integration: `HtmxHeaders`, partial page updates, `htmx_response_headers` |
-| `13_claude_agent_sdk` | Claude Agent SDK | Full agentic loop via `agenticapi-claude-agent-sdk` extension |
+| `13_claude_agent_sdk` | Claude Agent SDK | Full agentic loop via `agentharnessapi[claude-agent-sdk]` |
 | `14_dependency_injection` | None | `Depends()`, nested deps, `yield` teardown, `@tool` decorator |
 | `15_budget_policy` | None | `BudgetPolicy`, `PricingRegistry`, HTTP 402, spend tracking |
 | `16_observability` | None | OTEL tracing, Prometheus `/metrics`, `SqliteAuditRecorder` |
@@ -497,7 +494,7 @@ See [examples/README.md](examples/README.md) for the full examples guide.
 
 ### Exposing Endpoints via MCP
 
-1. Install: `pip install agenticapi[mcp]`
+1. Install: `pip install agentharnessapi[mcp]`
 2. Mark endpoints: `@app.agent_endpoint(name="search", enable_mcp=True)`
 3. Mount: `expose_as_mcp(app)`
 4. Test: `npx @modelcontextprotocol/inspector http://localhost:8000/mcp`
@@ -632,7 +629,7 @@ Extensions live under `extensions/<package-name>/` with their own `pyproject.tom
 3. Use **lazy imports** for the wrapped library so `import my_extension` never fails even when the optional dep is absent. Raise a friendly `*NotInstalledError` on first use.
 4. Tests must run **offline**: install a stub module in `conftest.py` that mimics the wrapped library's public surface.
 5. Errors should inherit from `agenticapi.AgenticAPIError` so callers can catch both core and extension errors uniformly.
-6. Reference: `extensions/agenticapi-claude-agent-sdk/` and `docs/internals/extensions.md`.
+6. Reference: `src/agenticapi/ext/claude_agent_sdk/` and `docs/internals/extensions.md`.
 
 ## Forward Tracks — Implementation Guide
 

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/CONTRIBUTING.md

@@ -119,7 +119,7 @@ $ uv run pytest tests/benchmarks/
 Extensions have their own test suites that run offline (no network, no API keys):
 
 ```bash
-$ uv run pytest extensions/agenticapi-claude-agent-sdk/tests
+$ uv run pytest tests/unit/ext/claude_agent_sdk/
 ```
 
 ---

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentharnessapi
-Version: 0.1.0
+Version: 0.1.1.dev2
 Summary: Agent-native web framework with harness engineering for Python
 Project-URL: Homepage, https://github.com/shibuiwilliam/AgenticAPI
 Project-URL: Documentation, https://shibuiwilliam.github.io/AgenticAPI/
@@ -28,6 +28,8 @@ Requires-Dist: pydantic>=2.12.5
 Requires-Dist: python-multipart>=0.0.20
 Requires-Dist: starlette>=1.0.0
 Requires-Dist: structlog>=25.5.0
+Provides-Extra: claude-agent-sdk
+Requires-Dist: claude-agent-sdk<0.2,>=0.1.58; extra == 'claude-agent-sdk'
 Provides-Extra: mcp
 Requires-Dist: mcp>=1.27.0; extra == 'mcp'
 Description-Content-Type: text/markdown
@@ -132,6 +134,10 @@ Either way you get **27 runnable examples** to copy from, **1,310 passing tests*
 
 **Python 3.13+** is required. The framework uses `match`, `type` aliases, `StrEnum`, and other modern features.
 
+```bash
+pip install agentharnessapi
+```
+
 For development:
 
 ```bash
@@ -143,8 +149,8 @@ uv sync --group dev    # or: pip install -e ".[dev]"
 Optional extras:
 
 ```bash
-pip install agenticapi[mcp]                  # MCP server support
-pip install agenticapi-claude-agent-sdk      # Full Claude Agent SDK loop (separate package)
+pip install agentharnessapi[mcp]                  # MCP server support
+pip install agentharnessapi[claude-agent-sdk]      # Full Claude Agent SDK loop
 
 # Observability -- all optional, all graceful no-ops when missing
 pip install opentelemetry-api opentelemetry-sdk opentelemetry-exporter-otlp
@@ -350,7 +356,7 @@ If you know FastAPI, you already know the patterns:
 | **Agent-to-Agent** | Capability discovery, trust scoring, inter-agent communication |
 | **Sessions** | Multi-turn conversations with context accumulation and TTL expiration |
 | **REST compatibility** | Mount FastAPI inside AgenticAPI, or expose agent endpoints as REST routes |
-| **Extensions** | Independent packages like `agenticapi-claude-agent-sdk` for heavyweight integrations |
+| **Extensions** | Optional extras like `agentharnessapi[claude-agent-sdk]` for heavyweight integrations |
 | **Observability** | OpenTelemetry spans + Prometheus metrics + W3C trace propagation, graceful no-op when absent |
 | **Eval harness** | Regression-test agent endpoints with deterministic assertion suites |
 | **OpenAPI docs** | Auto-generated Swagger UI, ReDoc, and OpenAPI 3.1.0 schema |
@@ -597,7 +603,7 @@ File upload via multipart, download via `FileResult`, streaming via Starlette. S
 ## MCP, REST Compatibility & Middleware
 
 ```python
-# Expose endpoints as MCP tools (pip install agenticapi[mcp])
+# Expose endpoints as MCP tools (pip install agentharnessapi[mcp])
 from agenticapi.interface.compat.mcp import expose_as_mcp
 expose_as_mcp(app, path="/mcp")
 
@@ -632,11 +638,11 @@ W3C trace propagation, request/latency/cost/denial metrics, graceful no-ops when
 Heavyweight integrations are released as separate packages under `extensions/`:
 
 ```bash
-pip install agenticapi-claude-agent-sdk
+pip install agentharnessapi[claude-agent-sdk]
 ```
 
 ```python
-from agenticapi_claude_agent_sdk import ClaudeAgentRunner
+from agenticapi.ext.claude_agent_sdk import ClaudeAgentRunner
 
 runner = ClaudeAgentRunner(
     system_prompt="You are a coding assistant.",
@@ -798,7 +804,7 @@ src/agenticapi/
     cli/                    # dev, console, replay, eval, init, version
 
 extensions/
-    agenticapi-claude-agent-sdk/   # Independent package -- full Claude Agent SDK loop
+    agenticapi-claude-agent-sdk/   # Historical — now merged into agentharnessapi[claude-agent-sdk]
 
 examples/
     01_hello_agent/ .. 27_multi_agent_pipeline/   # 27 runnable example apps

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/PROJECT.md

@@ -572,8 +572,8 @@ AgenticAPI/
 ├── src/agenticapi/        ← Framework source
 ├── tests/                 ← unit / integration / e2e / benchmarks
 ├── examples/              ← 27 runnable example apps (01 → 27)
-├── extensions/            ← Separately-installable extensions
-│     └── agenticapi-claude-agent-sdk/
+├── extensions/            ← Historical (now merged as optional extras)
+│     └── agenticapi-claude-agent-sdk/  (use: pip install agentharnessapi[claude-agent-sdk])
 │
 ├── docs/                  ← mkdocs site (served at /docs URL)
 │   ├── index.md           ← Docs hub

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/README.md

@@ -98,6 +98,10 @@ Either way you get **27 runnable examples** to copy from, **1,310 passing tests*
 
 **Python 3.13+** is required. The framework uses `match`, `type` aliases, `StrEnum`, and other modern features.
 
+```bash
+pip install agentharnessapi
+```
+
 For development:
 
 ```bash
@@ -109,8 +113,8 @@ uv sync --group dev    # or: pip install -e ".[dev]"
 Optional extras:
 
 ```bash
-pip install agenticapi[mcp]                  # MCP server support
-pip install agenticapi-claude-agent-sdk      # Full Claude Agent SDK loop (separate package)
+pip install agentharnessapi[mcp]                  # MCP server support
+pip install agentharnessapi[claude-agent-sdk]      # Full Claude Agent SDK loop
 
 # Observability -- all optional, all graceful no-ops when missing
 pip install opentelemetry-api opentelemetry-sdk opentelemetry-exporter-otlp
@@ -316,7 +320,7 @@ If you know FastAPI, you already know the patterns:
 | **Agent-to-Agent** | Capability discovery, trust scoring, inter-agent communication |
 | **Sessions** | Multi-turn conversations with context accumulation and TTL expiration |
 | **REST compatibility** | Mount FastAPI inside AgenticAPI, or expose agent endpoints as REST routes |
-| **Extensions** | Independent packages like `agenticapi-claude-agent-sdk` for heavyweight integrations |
+| **Extensions** | Optional extras like `agentharnessapi[claude-agent-sdk]` for heavyweight integrations |
 | **Observability** | OpenTelemetry spans + Prometheus metrics + W3C trace propagation, graceful no-op when absent |
 | **Eval harness** | Regression-test agent endpoints with deterministic assertion suites |
 | **OpenAPI docs** | Auto-generated Swagger UI, ReDoc, and OpenAPI 3.1.0 schema |
@@ -563,7 +567,7 @@ File upload via multipart, download via `FileResult`, streaming via Starlette. S
 ## MCP, REST Compatibility & Middleware
 
 ```python
-# Expose endpoints as MCP tools (pip install agenticapi[mcp])
+# Expose endpoints as MCP tools (pip install agentharnessapi[mcp])
 from agenticapi.interface.compat.mcp import expose_as_mcp
 expose_as_mcp(app, path="/mcp")
 
@@ -598,11 +602,11 @@ W3C trace propagation, request/latency/cost/denial metrics, graceful no-ops when
 Heavyweight integrations are released as separate packages under `extensions/`:
 
 ```bash
-pip install agenticapi-claude-agent-sdk
+pip install agentharnessapi[claude-agent-sdk]
 ```
 
 ```python
-from agenticapi_claude_agent_sdk import ClaudeAgentRunner
+from agenticapi.ext.claude_agent_sdk import ClaudeAgentRunner
 
 runner = ClaudeAgentRunner(
     system_prompt="You are a coding assistant.",
@@ -764,7 +768,7 @@ src/agenticapi/
     cli/                    # dev, console, replay, eval, init, version
 
 extensions/
-    agenticapi-claude-agent-sdk/   # Independent package -- full Claude Agent SDK loop
+    agenticapi-claude-agent-sdk/   # Historical — now merged into agentharnessapi[claude-agent-sdk]
 
 examples/
     01_hello_agent/ .. 27_multi_agent_pipeline/   # 27 runnable example apps

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/ROADMAP.md

@@ -25,7 +25,7 @@
 - 1,304 main tests + 38 extension tests · mypy `--strict` clean · ruff format + lint clean · mkdocs `--strict` clean
 - 27 example apps (`examples/01_hello_agent` through `26_dynamic_pipeline`)
 - 75 symbols in `agenticapi.__all__`
-- 1 extension (`agenticapi-claude-agent-sdk` v0.1.0)
+- 1 optional extra (`agentharnessapi[claude-agent-sdk]`)
 
 ---
 

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/docs/getting-started/examples.md

@@ -80,12 +80,12 @@ agenticapi dev --app examples.07_comprehensive.app:app
 
 **Demonstrates:** Multi-feature composition per endpoint: pipeline + A2A trust + multi-tool + approval + audit + sessions in each handler.
 
-## 08 — MCP Agent (requires `pip install agenticapi[mcp]`)
+## 08 — MCP Agent (requires `pip install agentharnessapi[mcp]`)
 
 Task tracker exposing select endpoints as MCP tools via the Model Context Protocol.
 
 ```bash
-pip install agenticapi[mcp]
+pip install agentharnessapi[mcp]
 uvicorn examples.08_mcp_agent.app:app --reload
 # Test MCP with the inspector:
 npx @modelcontextprotocol/inspector http://127.0.0.1:8000/mcp
@@ -155,13 +155,13 @@ curl -X POST http://127.0.0.1:8000/agent/todo.add -H "Content-Type: application/
 
 **Demonstrates:** `HtmxHeaders` auto-injection, `HTMLResult` for fragments and full pages, `htmx_response_headers()` for client-side control, conditional rendering based on `htmx.is_htmx`.
 
-## 13 — Claude Agent SDK (requires `agenticapi-claude-agent-sdk` extension)
+## 13 — Claude Agent SDK (requires `agentharnessapi[claude-agent-sdk]`)
 
 Runs a full Claude Agent SDK planning + tool-use loop inside an agent endpoint, with
 AgenticAPI policies bridged into the SDK permission system and audit trails preserved.
 
 ```bash
-pip install agenticapi-claude-agent-sdk
+pip install agentharnessapi[claude-agent-sdk]
 export ANTHROPIC_API_KEY="sk-ant-..."
 uvicorn examples.13_claude_agent_sdk.app:app --reload
 

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/docs/getting-started/installation.md

@@ -5,7 +5,13 @@
 - Python >= 3.13
 - pip or uv
 
-## Install from Source
+## Install from PyPI
+
+```bash
+pip install agentharnessapi
+```
+
+## Install from Source (development)
 
 ```bash
 git clone https://github.com/shibuiwilliam/AgenticAPI.git
@@ -38,7 +44,7 @@ AgenticAPI installs the following core dependencies:
 
 ```bash
 # MCP support (lightweight optional extra)
-pip install agenticapi[mcp]
+pip install agentharnessapi[mcp]
 
 # OpenTelemetry tracing (no-op unless installed)
 pip install opentelemetry-api opentelemetry-sdk opentelemetry-exporter-otlp
@@ -55,7 +61,7 @@ Heavyweight integrations are released as separate packages:
 
 ```bash
 # Claude Agent SDK (full agentic loop with policies and audit)
-pip install agenticapi-claude-agent-sdk
+pip install agentharnessapi[claude-agent-sdk]
 ```
 
 See the [Extensions guide](../internals/extensions.md) for the full list and how to build your own.

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/docs/guides/architecture.md

@@ -206,7 +206,7 @@ Disable with `AgenticApp(docs_url=None, redoc_url=None, openapi_url=None)`.
 
 ## MCP (Model Context Protocol) Support
 
-Agent endpoints can be exposed as MCP tools for use by Claude Desktop, Cursor, and other MCP clients. Requires `pip install agenticapi[mcp]`.
+Agent endpoints can be exposed as MCP tools for use by Claude Desktop, Cursor, and other MCP clients. Requires `pip install agentharnessapi[mcp]`.
 
 ```python
 @app.agent_endpoint(name="search", enable_mcp=True)

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/docs/internals/claude-agent-sdk-extension-plan.md

@@ -1,7 +1,12 @@
 # Claude Agent SDK Extension — Implementation Plan
 
-**Status:** Approved (Phase 1 — initial release)
-**Target version:** `agenticapi-claude-agent-sdk` v0.1.0
+> **Note (post-migration).** This extension has been merged into the main
+> package as `agentharnessapi[claude-agent-sdk]`. Import from
+> `agenticapi.ext.claude_agent_sdk`. This document is preserved as
+> historical design rationale.
+
+**Status:** Shipped (merged into main package as optional extra)
+**Target version:** `agentharnessapi[claude-agent-sdk]` (was `agenticapi-claude-agent-sdk` v0.1.0)
 **Owner:** Core team
 **Author:** Implementation plan generated 2026-04-10
 
@@ -21,8 +26,8 @@ audit trails, approval workflows, and tool registries.
 A user can:
 
 ```bash
-pip install agenticapi
-pip install agenticapi-claude-agent-sdk
+pip install agentharnessapi
+pip install agentharnessapi-claude-agent-sdk
 ```
 
 ```python

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/docs/internals/extending.md

@@ -345,7 +345,7 @@ Both are no-ops unless `opentelemetry-api` is installed. See [observability.md](
 
 ## Exposing Endpoints via MCP
 
-1. Install: `pip install agenticapi[mcp]`
+1. Install: `pip install agentharnessapi[mcp]`
 2. Mark endpoints: `@app.agent_endpoint(name="x", enable_mcp=True)`
 3. Mount: `expose_as_mcp(app)`
 4. Only `enable_mcp=True` endpoints become MCP tools

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/docs/internals/extensions.md

@@ -1,6 +1,14 @@
 # Extensions Architecture
 
-AgenticAPI supports independently-installable **extension packages** that wrap third-party libraries (LLM SDKs, framework adapters, domain-specific runtimes) without bloating the core package.
+> **Note (post-migration).** The Claude Agent SDK extension has been merged
+> into the main package as an optional extra. Install with
+> `pip install agentharnessapi[claude-agent-sdk]` and import from
+> `agenticapi.ext.claude_agent_sdk`. The `extensions/` directory is
+> preserved for historical reference. New extensions follow the same
+> extras pattern: add the dependency to `[project.optional-dependencies]`
+> in `pyproject.toml` and place the code under `src/agenticapi/ext/`.
+
+AgenticAPI supports **optional extras** that wrap third-party libraries (LLM SDKs, framework adapters, domain-specific runtimes) without bloating the core package.
 
 ## Why a separate-package layout?
 
@@ -160,8 +168,8 @@ List it in `pyproject.toml` if your build backend needs it declared explicitly.
 ### End users
 
 ```bash
-pip install agenticapi                          # Core
-pip install agenticapi-claude-agent-sdk         # Add extension (pulls its deps)
+pip install agentharnessapi                          # Core
+pip install agentharnessapi-claude-agent-sdk         # Add extension (pulls its deps)
 ```
 
 Extensions should never be required for core to work. Handlers that need an extension either import it explicitly at module load, or guard the import behind a runtime check.

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/docs/internals/modules.md

@@ -40,7 +40,7 @@ src/agenticapi/
         compat/
             rest.py          RESTCompat, expose_as_rest — REST route generation
             fastapi.py       mount_fastapi, mount_in_agenticapi — ASGI mount
-            mcp.py           MCPCompat, expose_as_mcp — MCP server (optional: pip install agenticapi[mcp])
+            mcp.py           MCPCompat, expose_as_mcp — MCP server (optional: pip install agentharnessapi[mcp])
         a2a/
             protocol.py      A2AMessage, A2AMessageType, A2ARequest, A2AResponse
             capability.py    Capability, CapabilityRegistry

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/examples/13_claude_agent_sdk/app.py

@@ -19,7 +19,7 @@ What this example shows:
   error response explaining how to install it.
 
 Prerequisites:
-    pip install agenticapi-claude-agent-sdk
+    pip install agentharnessapi[claude-agent-sdk]
     export ANTHROPIC_API_KEY=sk-ant-...
 
 Run with:
@@ -126,7 +126,7 @@ def _build_runner() -> tuple[Any, AuditRecorder] | None:
     call.
     """
     try:
-        from agenticapi_claude_agent_sdk import ClaudeAgentRunner
+        from agenticapi.ext.claude_agent_sdk import ClaudeAgentRunner
     except ImportError:
         return None
 
@@ -195,7 +195,8 @@ async def ask(intent: Intent, context: AgentContext) -> dict[str, Any]:
             "ok": False,
             "error": "extension_not_installed",
             "message": (
-                "agenticapi-claude-agent-sdk is not installed. Install it with: pip install agenticapi-claude-agent-sdk"
+                "agenticapi.ext.claude_agent_sdk is not available. "
+                "Install it with: pip install agentharnessapi[claude-agent-sdk]"
             ),
             "intent": intent.raw,
         }

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/examples/27_multi_agent_pipeline/app.py

@@ -4,7 +4,7 @@ Demonstrates a 3-role research pipeline: researcher → summariser → reviewer.
 All roles run in-process with budget propagation and trace linkage.
 
 Prerequisites:
-    pip install agenticapi
+    pip install agentharnessapi
 
 Run:
     agenticapi dev --app examples.27_multi_agent_pipeline.app:app

{agentharnessapi-0.1.0 → agentharnessapi-0.1.1.dev2}/examples/README.md

@@ -20,7 +20,7 @@ Every example automatically serves interactive API docs at `http://127.0.0.1:800
 | [10_file_handling](#10-file-handling) | Files | No | Upload: `UploadedFiles`, download: `FileResult`, streaming |
 | [11_html_responses](#11-html-responses) | Pages | No | `HTMLResult`, `PlainTextResult`, `FileResult`, mixed endpoints |
 | [12_htmx](#12-htmx) | Todo app | No | `HtmxHeaders`, `htmx_response_headers`, partial updates |
-| [13_claude_agent_sdk](#13-claude-agent-sdk) | Assistant + audit | `ANTHROPIC_API_KEY` (optional) | Full Claude Agent SDK loop via `agenticapi-claude-agent-sdk` extension |
+| [13_claude_agent_sdk](#13-claude-agent-sdk) | Assistant + audit | `ANTHROPIC_API_KEY` (optional) | Full Claude Agent SDK loop via `agentharnessapi[claude-agent-sdk]` |
 | [14_dependency_injection](#14-dependency-injection) | Bookstore | No | `Depends()`, nested dependencies, `yield` teardown, `@tool` decorator |
 | [15_budget_policy](#15-budget-policy) | Chat with cost caps | No | `BudgetPolicy`, `PricingRegistry`, HTTP 402 on budget breach, spend inspection |
 | [16_observability](#16-observability) | Production ops | No | `configure_tracing` / `configure_metrics`, `SqliteAuditRecorder`, Prometheus `/metrics` |
@@ -48,7 +48,7 @@ agenticapi dev --app examples.01_hello_agent.app:app
 uvicorn examples.01_hello_agent.app:app --reload
 ```
 
-Examples 01, 02, 08-12, and 14-24 require no API keys. Examples 03, 04, and 05 are designed for a specific LLM provider — they *import* cleanly without credentials and continue to serve `/health`, `/docs`, and their deterministic search / inventory / metrics endpoints, but the LLM-powered endpoints (LLM-driven code generation in 03, `products.describe` / `products.recommend` in 04, `tickets.analyze` / `tickets.draft_response` in 05) return a typed friendly error until the matching `OPENAI_API_KEY` / `ANTHROPIC_API_KEY` / `GOOGLE_API_KEY` is set. Examples 06 and 07 let you choose a provider via `AGENTICAPI_LLM_PROVIDER` and fall back to direct-handler mode when no key is set. Example 08 requires `pip install agenticapi[mcp]`. Example 13 requires `pip install agenticapi-claude-agent-sdk` and (for live calls) `ANTHROPIC_API_KEY` — without them it imports cleanly and the `assistant.audit` endpoint still works. Example 16 runs without OpenTelemetry installed (all tracing/metrics calls become no-ops) and upgrades itself when `opentelemetry-api` + `opentelemetry-sdk` are present. Examples 17 and 19 use `MockBackend` so the demo curl walkthroughs run without any LLM keys; swap in a real backend with a two-line change when you're ready to exercise the same code path against Anthropic, OpenAI, or Gemini.
+Examples 01, 02, 08-12, and 14-24 require no API keys. Examples 03, 04, and 05 are designed for a specific LLM provider — they *import* cleanly without credentials and continue to serve `/health`, `/docs`, and their deterministic search / inventory / metrics endpoints, but the LLM-powered endpoints (LLM-driven code generation in 03, `products.describe` / `products.recommend` in 04, `tickets.analyze` / `tickets.draft_response` in 05) return a typed friendly error until the matching `OPENAI_API_KEY` / `ANTHROPIC_API_KEY` / `GOOGLE_API_KEY` is set. Examples 06 and 07 let you choose a provider via `AGENTICAPI_LLM_PROVIDER` and fall back to direct-handler mode when no key is set. Example 08 requires `pip install agentharnessapi[mcp]`. Example 13 requires `pip install agentharnessapi[claude-agent-sdk]` and (for live calls) `ANTHROPIC_API_KEY` — without them it imports cleanly and the `assistant.audit` endpoint still works. Example 16 runs without OpenTelemetry installed (all tracing/metrics calls become no-ops) and upgrades itself when `opentelemetry-api` + `opentelemetry-sdk` are present. Examples 17 and 19 use `MockBackend` so the demo curl walkthroughs run without any LLM keys; swap in a real backend with a two-line change when you're ready to exercise the same code path against Anthropic, OpenAI, or Gemini.
 
 ---
 
@@ -410,7 +410,7 @@ A task tracker that exposes select endpoints as [MCP](https://modelcontextprotoc
 **Prerequisites:**
 
 ```bash
-pip install agenticapi[mcp]
+pip install agentharnessapi[mcp]
 ```
 
 ```bash
@@ -670,7 +670,7 @@ curl -X POST http://127.0.0.1:8000/agent/todo.toggle \
 
 ## 13 Claude Agent SDK
 
-A demo of the **`agenticapi-claude-agent-sdk`** extension, which runs the full Claude Agent SDK loop (planning + tool use + reflection) inside an AgenticAPI endpoint while preserving AgenticAPI's harness guarantees: declarative policies, an audit trail, and a tool registry exposed to the model as MCP tools.
+A demo of the **`agentharnessapi[claude-agent-sdk]`** extra, which runs the full Claude Agent SDK loop (planning + tool use + reflection) inside an AgenticAPI endpoint while preserving AgenticAPI's harness guarantees: declarative policies, an audit trail, and a tool registry exposed to the model as MCP tools.
 
 The example wires up a `ClaudeAgentRunner` with a `CodePolicy`, an in-process AgenticAPI tool (`FaqTool`), and an `AuditRecorder`. It also degrades gracefully when the extension or `ANTHROPIC_API_KEY` is missing — the app still imports, the `assistant.audit` endpoint still works, and `assistant.ask` returns a structured error explaining how to install the extension.
 
@@ -679,7 +679,7 @@ The example wires up a `ClaudeAgentRunner` with a `CodePolicy`, an in-process Ag
 **Prerequisites (optional but recommended):**
 
 ```bash
-pip install agenticapi-claude-agent-sdk
+pip install agentharnessapi[claude-agent-sdk]
 export ANTHROPIC_API_KEY="sk-ant-..."
 ```