agentic-systems 1.1.2__tar.gz → 2.0.0__tar.gz

agentic_systems-2.0.0/CHANGELOG.md

@@ -0,0 +1,224 @@
+# Changelog
+
+All notable changes to Agentic Systems are documented here.
+## 2.0.0 - 2026-08-19
+
+Stable release of the unified computation model. The stable API contains 78
+top-level exports, 370 traced export/member IDs, and 10 shared contract scenarios.
+
+### Added
+
+- `Executable`, `ExecutionPlan`, `SequentialPlan`, and `ParallelPlan` as the
+  common contract for tools, agents, pipelines, and systems.
+- Hierarchical `RunResult` values with execution and parent identity, child
+  results, and tree traversal.
+- `ToolSet` as the public name for a namespaced collection of tools.
+- Pure `toolkit.skill(path=...)` loading without a hidden `AgenticSystem`.
+- `ModelProviderConfig` and `toolkit.provider(...)`, separating provider
+  configuration from runtime execution.
+- `Agent.pipeline(...)`, `AgenticSystem.add(...)`, `compile(...)`, and direct
+  system `run`/`arun` execution.
+- Generic `Evaluator.evaluate(...)` support for any executable agent or system.
+- A public 5 Provider x 4 Framework compatibility report and matrix notebook
+  covering all 20 combinations.
+- Portable `vllm`/`vllm-client` extras and a separate `vllm-server` extra.
+- Real Native, LangGraph, OpenAI Agents, and Strands adapter dispatch.
+- FrameworkConfig, exact native kwargs forwarding, Agent.prepare,
+  Agent.native_agent, and RunResult.native_result.
+- Offline 5 Provider x 4 Framework certification with fake transports and real
+  Framework SDK loops.
+- Layered 21-notebook Python curriculum mirrored by 21 preserved-output CLI notebooks.
+- Deterministic Framework and matrix tutorials executed from clean kernels.
+- OpenAI Agents offline certification for mixed Tools, typed output, sessions, guardrails and handoffs.
+- Strands offline certification for mixed Tools, hooks, structured output and sync/async execution.
+- Native Strands MCP certification over local stdio and Streamable HTTP
+  transports, executed through the deterministic Python Provider.
+- Non-secret Provider environment snapshots now load the nearest `.env`
+  consistently with runtime factories and the CLI; OpenAI gains the symmetric
+  `openai_environment_snapshot()` public helper.
+- OpenAI Agents tracing is disabled per run by default for non-OpenAI Providers,
+  preventing cross-provider trace export when several credentials coexist.
+- Native Bedrock API-key recognition across auto resolution, diagnostics, and
+  notebook readiness. `AWS_BEARER_TOKEN_BEDROCK` and the standard AWS credential
+  chain are two authentication modes of the same boto3 `bedrock-runtime`
+  Provider; boto3/botocore >=1.39 is required for the Bearer mode.
+- Agentic Systems Studio as a portable product bundle with ten independently
+  reusable system bundles, a system creator, SQLite catalog, Mermaid topology,
+  notebooks, CLI, tests and sandbox validation.
+- A standalone, reproducible `agentic-systems-skill-2.0.0.zip` skill artifact.
+
+### Removed
+
+- providers.python_direct, engines.python_direct, PYTHON_DIRECT_ENGINE,
+  and the tools.compat quarantine.
+- Runtime engine coercion, include_aliases, and output_contains aliases.
+- style-only and declarative-only Framework states.
+- The Bedrock-specific OpenAI Agents bridge, its nine public helper methods and
+  its `disable_openai_runtime_tracing` constructor option. OpenAI Agents over
+  Bedrock now uses the general Framework adapter exclusively.
+- All disable_framework_tracing parameters. 2.0 has no parallel public switch
+  or shim; native SDK configuration remains available, with a safe no-egress
+  default for OpenAI Agents runs backed by non-OpenAI Providers.
+- The domain-specific `tutorials/skills/accountability_otc` legacy tutorial and
+  its private-data-shaped tests. The neutral `tutorial_api_inspection` filesystem
+  Skill now teaches the 2.0 runtime Skill contract; OTC assets are absent from
+  release bundles.
+
+The tutorial gate executes 17 deterministic notebooks, validates 4 Provider notebooks
+offline, and verifies 21 preserved-output CLI notebooks mapped 1:1 to the Python curriculum.
+
+Live OpenAI and local Ollama `qwen3:4b` evidence passed all four Frameworks
+through the strict matrix gate with real tool calls and normalized `RunResult`
+values; Ollama executed 100% on an NVIDIA GPU. Bedrock bearer authentication and
+inference-profile discovery passed, but live agent execution is not certified
+because the account returned its daily-token `ThrottlingException`. Bedrock IAM
+and vLLM server evidence remain external until their environments execute the
+same `--live --require-pass` route.
+The Bedrock ratchet rises to 100% over all 620 remaining statements.
+Branch ratchets are fixed at Core 98.1%, Providers 97.7%, and Frameworks 98.8%;
+each threshold is the measured alpha result truncated to one decimal and may
+only increase.
+
+
+## 1.1.3 - 2026-08-11
+
+Documentation-maintenance release preserving the 1.1 runtime behavior and
+111-symbol public API while consolidating current product evidence.
+
+### Changed
+
+- Consolidated conceptual, semantic and contract documents into explicit current
+  sources of truth.
+- Removed repository-local ADRs, an unaccepted RFC, migration notes and
+  release/checkpoint narratives after consolidating every still-current
+  contract into the canonical product documentation.
+- Corrected stale Graph backend, migration, CLI inventory and release narratives.
+- Reduced `docs/` to current product guidance; Git, this changelog and GitHub
+  Releases remain the historical evidence layers.
+- Reorganized the 66-module pytest layout into 92 owner-based modules across
+  API, unit, integration, Provider, and release layers; removed historical
+  `extended`, `branches`, `residual`, and `remaining` naming from test files and
+  test functions.
+- Removed one redundant negative wrapper-absence test. The resulting 383 tests
+  preserve the behavioral boundary through the frozen public-API and test-suite
+  architecture gates.
+- Made API/documentation coherence executable: the API index must equal the 111
+  top-level symbols in order, every documented `toolkit.*` call is checked against
+  source signatures, and each tutorial `api_coverage` inventory must equal its real
+  AST usage.
+
+## 1.1.2 - 2026-08-10
+
+Maintenance release completing the controlled retirement of legacy tests and
+the internal partition of the Bedrock runtime.
+
+### Changed
+
+- Migrated or retired all 52 routed legacy test modules into owner-based API,
+  unit, Provider, release, and contract suites; removed the quarantine, dynamic
+  loader, Ruff exclusion, and numeric inventory guard.
+- Split the 2,961-line Bedrock implementation behind its unchanged public
+  `providers.bedrock_runtime` facade into focused internal components; reduced
+  `run_direct` to a 37-line orchestrator.
+- Added frozen public-API and Bedrock-signature contracts plus Python 3.10/3.14
+  CI gates for Ruff, tests, coverage, packaging, and isolated wheel smoke.
+- Kept core coverage at 100% over 6,194 statements and introduced a separate
+  Bedrock coverage ratchet with `fail_under = 53.1` (measured 53.17%).
+- The tutorial gate executes 13 deterministic notebooks.
+- It checks 5 Provider notebooks statically; live OpenAI, Bedrock and vLLM
+  execution remains outside the release claim.
+- Removed four genuinely unused imports from active source.
+## 1.1.1 - Unpublished checkpoint (2026-08-10)
+
+Unpublished maintenance checkpoint focused on coherence and surface reduction; superseded by 1.1.2.
+
+### Changed
+
+- Removed accidental top-level `build_single_agent_step_graph` and `PUBLIC_API`
+  attributes; the canonical routes are `toolkit.graph(...)` and `toolkit.__all__`.
+- Removed the undocumented `vll` extra and the misleading `tutorials` extra.
+  Notebooks remain repository content and their dependencies now live in `dev`.
+- Scoped the 100% coverage claim to core modules because the Bedrock runtime is
+  explicitly omitted from the coverage configuration.
+- Separated deterministic notebook execution evidence from static Provider
+  notebook checks.
+- Updated stable installation and onboarding instructions.
+
+## 1.1.0 - 2026-07-19
+
+Stable 1.1 release. The release candidate was promoted after the complete
+manual notebook matrix and updated automated gates passed.
+
+### Closure evidence
+
+- 18 of 18 canonical notebooks executed from fresh kernels; no failures.
+- 393 tests passed with 100.00% coverage over 6,193 statements.
+- Canonical construction remains under `import agentic_systems as toolkit`.
+- Added ergonomic `toolkit.skill(...)`, `toolkit.environment(...)`, and
+  `toolkit.eval()` factories over existing public types.
+- Preserved `toolkit.system(...)` to avoid shadowing the supported
+  `agentic_systems.system` module import.
+- vLLM installation, server startup, SDK smoke and Agentic Systems smoke are
+  opt-in and report explicit skips when disabled.
+- Live OpenAI, Bedrock and vLLM execution remains outside the release claim.
+
+## 1.1.0rc1 - 2026-07-18
+
+Release candidate. Full notebook execution remains a manual release gate.
+
+### Added
+
+- Normative computational grammar and semantics.
+- RunResult invariants for success, validation, partial failure, errors, usage,
+  tool events, lineage, evidence, and serialization.
+- Explicit Tool and Skill identity, conflict, precedence, and composition laws.
+- Provider capability profiles and a shared conformance suite.
+- Framework and Graph boundary profiles.
+- Environment/Eval ownership and reproducibility classifications.
+- Public `InspectReport` with structured and stable human projections.
+- Static System diagnostics for relationships, contracts, capabilities,
+  conflicts, limits, and degradation risks.
+- Migration guide from 1.0 to 1.1.
+
+### Changed
+
+- The canonical deterministic Provider name taught by docs is
+  `python-runtime`; `python-direct` remains a compatibility alias.
+- Tool and Skill collisions require an explicit decision.
+- Provider and Framework concepts are documented independently.
+- Auto Provider resolution requires both AWS authentication and region signals
+  before selecting Bedrock; a region alone no longer masks configured OpenAI or
+  vLLM backends.
+- Tutorial LM reviewers degrade explicitly when optional Provider execution
+  fails, while required deterministic results preserve their own status.
+- Arithmetic integration tutorials validate expected tool outputs, preventing a
+  final answer from contradicting its execution evidence.
+- Tutorials 11-14 were consolidated into three end-to-end paths: single
+  AgenticSystem, multi-agent System, and the same multi-agent System as Graph.
+- Eval reports record deterministic, seeded, or non-deterministic replay
+  conditions.
+- The public API grows additively from 105 to 106 symbols with
+  `InspectReport`.
+
+### Compatibility
+
+- `import agentic_systems as toolkit` is unchanged.
+- Tool and Agent execution continues to return `RunResult`.
+- `AgenticSystem.inspect()` remains dictionary-compatible.
+- No public `ExecutionContext` object was added.
+- No new optional runtime dependency is required by the core package.
+
+### Verified Evidence
+
+- Unit, contract, composition, and integration-boundary tests.
+- Static compilation and source-clean checks for all 18 canonical notebooks.
+- Package API inventory and documentation checksum.
+- Wheel and source distribution content inspection.
+
+Live OpenAI, Bedrock, vLLM, Strands SDK, and OpenAI Agents SDK execution are not
+claimed. Strands and OpenAI Agents SDK adapters are not part of 1.1.
+
+## 1.0.7
+
+Baseline stable release used by the 1.1 grammar audit. Historical details remain
+in the repository history and checkpoint 1.1.0 audit.

{agentic_systems-1.1.2/src/agentic_systems.egg-info → agentic_systems-2.0.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentic-systems
-Version: 1.1.2
+Version: 2.0.0
 Summary: Agentic Systems: a computational grammar for building, executing, observing, and evaluating intelligent systems.
 Author-email: Jacobo Gerardo Gonzalez Leon <jacobogerardo.gonzalez@bbva.com>
 License-Expression: MIT
@@ -21,16 +21,25 @@ Requires-Dist: pandas>=2.0.0
 Requires-Dist: typing-extensions>=4.10.0
 Requires-Dist: rich>=13.0.0
 Provides-Extra: bedrock
-Requires-Dist: boto3>=1.34.0; extra == "bedrock"
-Requires-Dist: botocore>=1.34.0; extra == "bedrock"
+Requires-Dist: boto3>=1.39.0; extra == "bedrock"
+Requires-Dist: botocore>=1.39.0; extra == "bedrock"
 Provides-Extra: langgraph
 Requires-Dist: langgraph>=0.2.0; extra == "langgraph"
 Requires-Dist: typing-extensions>=4.10.0; extra == "langgraph"
 Provides-Extra: openai
-Requires-Dist: openai>=1.0.0; extra == "openai"
+Requires-Dist: openai<3,>=2.45.0; extra == "openai"
+Provides-Extra: openai-agents
+Requires-Dist: openai-agents<0.19,>=0.18.3; extra == "openai-agents"
+Provides-Extra: strands
+Requires-Dist: strands-agents<2,>=1.29.0; extra == "strands"
+Requires-Dist: mcp<2,>=1; extra == "strands"
 Provides-Extra: vllm
-Requires-Dist: vllm>=0.9.0; extra == "vllm"
-Requires-Dist: openai>=1.0.0; extra == "vllm"
+Requires-Dist: openai<3,>=2.45.0; extra == "vllm"
+Provides-Extra: vllm-client
+Requires-Dist: openai<3,>=2.45.0; extra == "vllm-client"
+Provides-Extra: vllm-server
+Requires-Dist: vllm>=0.9.0; extra == "vllm-server"
+Requires-Dist: openai<3,>=2.45.0; extra == "vllm-server"
 Provides-Extra: dev
 Requires-Dist: pytest>=8.0.0; extra == "dev"
 Requires-Dist: pytest-cov>=5.0.0; extra == "dev"
@@ -41,11 +50,13 @@ Requires-Dist: jupyterlab>=4.0.0; extra == "dev"
 Requires-Dist: nbclient>=0.10.0; extra == "dev"
 Requires-Dist: nbformat>=5.10.0; extra == "dev"
 Provides-Extra: all
-Requires-Dist: boto3>=1.34.0; extra == "all"
-Requires-Dist: botocore>=1.34.0; extra == "all"
+Requires-Dist: boto3>=1.39.0; extra == "all"
+Requires-Dist: botocore>=1.39.0; extra == "all"
 Requires-Dist: langgraph>=0.2.0; extra == "all"
-Requires-Dist: openai>=1.0.0; extra == "all"
-Requires-Dist: vllm>=0.9.0; extra == "all"
+Requires-Dist: openai<3,>=2.45.0; extra == "all"
+Requires-Dist: openai-agents<0.19,>=0.18.3; extra == "all"
+Requires-Dist: strands-agents<2,>=1.29.0; extra == "all"
+Requires-Dist: mcp<2,>=1; extra == "all"
 Requires-Dist: ipykernel>=6.0.0; extra == "all"
 Requires-Dist: jupyterlab>=4.0.0; extra == "all"
 Requires-Dist: nbclient>=0.10.0; extra == "all"
@@ -79,7 +90,7 @@ Its Python API turns that grammar into explicit, composable abstractions: tools,
 
 Use it when agentic workloads need to be observable, testable, portable and ready for repeated execution, not just notebook demos.
 
-Agentic Systems 1.1 keeps explicit traceability between its API,
+Agentic Systems 2.0 keeps explicit traceability between its API,
 documentation, tutorials, and tests.
 
 ```text
@@ -88,7 +99,7 @@ API -> Docs -> Tutorials -> explicit automated or manual evidence
 
 Public concepts are defined in the API, explained in the documentation, taught through the canonical tutorials, and checked by explicit release gates.
 
-Release status: `1.1.2` completes test-infrastructure cleanup and partitions Bedrock while preserving the 111-symbol public API.
+Release status: `2.0.0` is the Provider x Framework architecture: 78 stable top-level exports, 370 traced export/member IDs, 10 shared contract scenarios, four real Framework adapters, and no silent fallback.
 
 ## Installation
 
@@ -100,6 +111,31 @@ pip install agentic-systems
 import agentic_systems as toolkit
 ```
 
+### Install the Agentic Systems skill
+
+The `v2.0.0` GitHub release includes a credential-free skill ZIP whose archive
+root is `agentic-systems/`.
+
+```powershell
+Expand-Archive .\agentic-systems-skill-2.0.0.zip `
+  -DestinationPath "$env:USERPROFILE\.codex\skills" -Force
+```
+
+```bash
+unzip agentic-systems-skill-2.0.0.zip -d ~/.codex/skills
+```
+
+Restart or reload Codex, then invoke `$agentic-systems`. OpenAI Skills upload
+surfaces that accept a ZIP can consume the same artifact. Download the
+standalone skill and the complete ten-system Studio from the
+[v2.0.0 release](https://github.com/JacoboGGLeon/agentic_systems/releases/tag/v2.0.0).
+
+Release assets:
+
+- `agentic-systems-skill-2.0.0.zip`: Codex/OpenAI skill package.
+- `agentic-systems-studio-2.0.0.zip`: portable Studio with ten nested systems.
+- `SHA256SUMS-2.0.0.txt`: hashes for the Python and product artifacts.
+
 For an installed-package smoke test and provider notebook setup, follow
 [First Run Onboarding](docs/ONBOARDING_FIRST_RUN.md).
 
@@ -128,7 +164,7 @@ Agentic Systems represents those concerns as a computational grammar: runtime, t
 | Environments | Episodic execution over records, transitions, rewards and history. |
 | Evals | Repeatable validation cases with pass/fail reporting. |
 | Lineage Memory | Human-readable explanation of what happened and why. |
-| Integrations | A portable Graph backend, an optional LangGraph adapter, and explicit framework identities. |
+| Integrations | Native, LangGraph, OpenAI Agents, and Strands execution over explicit Providers. |
 
 ## Core Model
 
@@ -161,17 +197,17 @@ Canonical providers:
 | `python-runtime` | Local deterministic execution for tools, policies and smoke tests. |
 | `auto` | Selects a concrete provider from environment signals before execution. |
 
-Default `provider="auto"` priority is `bedrock-runtime`, then `openai-runtime`, then `vllm-runtime`. Override it with `provider_priority=[...]` or `AGENTIC_SYSTEMS_PROVIDER_PRIORITY=...`. Bedrock is considered configured only when both a region and an AWS authentication signal are present; a region alone does not outrank a usable OpenAI or vLLM configuration.
+Default `provider="auto"` priority is `bedrock-runtime`, then `openai-runtime`, `vllm-runtime`, and `ollama-runtime`. Override it with `provider_priority=[...]` or `AGENTIC_SYSTEMS_PROVIDER_PRIORITY=...`. Bedrock is considered configured only when both a region and an AWS authentication signal are present. That signal may be the standard AWS credential chain (signed with SigV4) or the native `AWS_BEARER_TOKEN_BEDROCK` API key; both use the same boto3 `bedrock-runtime` Provider and a region alone does not outrank another usable Provider.
 
 Canonical framework facades:
 
 | Framework | Use |
 |---|---|
 | `langgraph` | LangGraph graph orchestration. |
-| `openai-agents` | Style-only identity over the selected runtime; no OpenAI Agents SDK adapter. |
-| `strands` | Declarative compatibility identity; no Strands SDK adapter. |
+| `openai-agents` | OpenAI Agents `Runner`, native Tools, handoffs, guardrails, sessions and MCP. |
+| `strands` | `strands.Agent`, native Tools/MCP, hooks and lifecycle. |
 
-Providers decide where execution runs. A real Framework adapter may own the outer orchestration loop; an accepted framework label alone does not prove adapter execution. Configuration details live in `docs/ONBOARDING_FIRST_RUN.md`, `docs/CLI.md` and the `tutorials/00_runtime_*` notebooks.
+Providers decide where inference runs. The selected Framework always owns its orchestration loop, and `RunResult.meta["framework_adapter"]` records the adapter that executed. Configuration details live in docs/ONBOARDING_FIRST_RUN.md, docs/CLI.md and the layered tutorials/core, tutorials/providers and tutorials/frameworks paths.
 
 ## From Zero-to-Hero
 
@@ -402,61 +438,51 @@ The CLI is for inspection, diagnostics and packaging smoke tests. It should not
 
 ## Tutorials
 
-The official learning path is `tutorials/`. It explains and exercises the public API directly from `import agentic_systems as toolkit`.
+The canonical learning path is layered by responsibility:
+
+| Layer | Purpose | Default execution |
+|---|---|---|
+| tutorials/core | Agentic Systems grammar and composition | Python Runtime, Native Framework, portable Graph |
+| tutorials/providers | Inference boundary | Explicit preflight; external execution when configured |
+| tutorials/frameworks | Native orchestration SDK | Real SDK offline; Provider auto is optional |
+| tutorials/api | Exact public contract | Generated manifest, documentation and CLI/Pytest traceability |
 
-Provider notebooks require no activation cell. Configure the external boundary,
-open the corresponding notebook, and choose **Run All**:
+Provider notebooks use one public route and choose Run All:
 
 | Provider | Readiness inputs | Notebook |
 |---|---|---|
-| OpenAI | `OPENAI_API_KEY`; optional `OPENAI_MODEL` | `00_runtime_openai_provider_api.ipynb` |
-| vLLM | `VLLM_BASE_URL` and `VLLM_MODEL` | `00_runtime_vllm_provider_api.ipynb` |
-| Bedrock | Standard AWS credential chain; optional model/region overrides | `00_runtime_bedrock_provider_api.ipynb` |
+| OpenAI | OPENAI_API_KEY; OPENAI_MODEL optional | tutorials/providers/01_openai.ipynb |
+| vLLM | VLLM_BASE_URL and VLLM_MODEL | tutorials/providers/03_vllm.ipynb |
+| Bedrock | Region plus SigV4 credentials or AWS_BEARER_TOKEN_BEDROCK | tutorials/providers/02_bedrock.ipynb |
+| Ollama | OLLAMA_MODEL; OLLAMA_BASE_URL optional | tutorials/providers/04_ollama.ipynb |
 
-Live execution is the default when readiness passes. A provider-specific
-`RUN_*_LIVE=0` is an explicit opt-out, not a prerequisite.
-
-```text
-tutorials/00_runtime_api.ipynb
-tutorials/00_runtime_bedrock_provider_api.ipynb
-tutorials/00_runtime_openai_provider_api.ipynb
-tutorials/00_runtime_scheduler_api.ipynb
-tutorials/00_runtime_vllm_provider_api.ipynb
-tutorials/01_tool_api.ipynb
-tutorials/02_skill_api.ipynb
-tutorials/03_agent_api.ipynb
-tutorials/04_human_result_api.ipynb
-tutorials/05_lineage_memory_api.ipynb
-tutorials/06_integrations_strands_api.ipynb
-tutorials/07_integrations_openai_runtime_api.ipynb
-tutorials/08_system_api.ipynb
-tutorials/09_graph_api.ipynb
-tutorials/10_environment_eval_api.ipynb
-tutorials/11_single_agentic_system_api.ipynb
-tutorials/12_multi_agentic_system_api.ipynb
-tutorials/13_multi_agentic_graph_api.ipynb
-```
+The Framework notebooks execute LangGraph, OpenAI Agents and Strands SDKs for
+real with python-runtime; the matrix notebook covers all 20 declared pairs.
+Optional RUN_*_LIVE=1 changes only the Provider.
+RUN_*_LIVE=0 forces a fully offline validation.
 
-There is no active `examples/` root. Tutorials are the executable documentation.
+The full ordered inventory of 21 Python notebooks and their 21 CLI mirrors lives in
+tutorials/README.md. There is no active examples/ root or duplicate notebook
+route.
 
 ## Documentation
 
 Use the [documentation map](docs/README.md) to choose the user guide,
-conceptual model, API/CLI reference, engineering gates or historical evidence.
-The current release evidence is recorded in
-[Agentic Systems 1.1.2](docs/RELEASE_1_1_2.md).
+conceptual model, API/CLI reference or current engineering contracts. Version
+history is recorded in the [changelog](CHANGELOG.md); published artifact
+evidence belongs to GitHub Releases.
 
 ## Quality Gate
 
 Current verified status:
 
 ```text
-Version: 1.1.2
+Version: 2.0.0
 PyPI package: agentic-systems
 Tests: run `python -m pytest` for the current count
 Core coverage: 100.00%
-Coverage scope: Bedrock facade and internal package excluded from core; separately gated at 53.1%
-Canonical notebooks: 13 deterministic executed; 5 Provider notebooks checked statically
+Coverage scope: Bedrock facade and internal package excluded from core; separately gated at 100%
+Canonical notebooks: 17 deterministic executed; 3 live Provider notebooks checked statically
 
 ```