anycode-py 0.6.0__tar.gz → 0.7.0__tar.gz

{anycode_py-0.6.0 → anycode_py-0.7.0}/.gitignore

@@ -4,6 +4,7 @@ __pycache__/
 *.egg-info/
 dist/
 build/
+site/
 .eggs/
 *.egg
 .venv/

{anycode_py-0.6.0 → anycode_py-0.7.0}/CHANGELOG.md

@@ -7,6 +7,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.7.0] - 2026-07-11
+
+### Added
+
+- **MCP, plugin, and tool trust hardening** - agent tool allowlists are now enforced again at execution, including an explicit empty list, so a provider cannot invoke an unadvertised registered tool. MCP tools require exact per-agent server opt-in, remain bound to discovery ownership across prebuilt agents and reconnects, always use side-effect idempotency, fail closed when configured auth is missing, and clean up partial initialization without masking cancellation. Plugin entry points are filtered before import and plugin contributions are preflighted before shared registry mutation. The security threat model and production-readiness checklist define the remaining host, network, identity, storage, and operational responsibilities.
+- **Cross-platform compatibility CI** - pull requests and `main` now run the complete non-integration suite on Linux, Windows, and macOS across Python 3.12 and 3.13. Separate jobs enforce locked quality checks, core-only and per-extra dependency isolation, Redis/ChromaDB integration coverage, and built wheel smoke tests for both core and CLI installations. Repository tests keep the optional-extra matrix synchronized with package metadata.
+- **Enforced compatibility contracts** - the complete v0.6 top-level Python API now has an additive CI baseline, and duplicate public declarations fail tests. Declarative YAML/TOML files use format v1, preserve unversioned v1 compatibility, and reject future versions with `UnsupportedConfigVersionError`. The compatibility reference defines public import boundaries, semantic-version rules, checkpoint and durable-run reader ranges, plus upgrade and rollback procedures.
+- **Operational observability** - agent runs now share durable run/trace correlation across turn, LLM, tool, and terminal spans, with task-local async parenting and deterministic per-trace sampling. Completed spans automatically feed bounded latency, first-token, token, estimated-cost, retry, outcome, and error metrics plus redacted structured events. The new `jsonl` exporter emits one correlated completion record per sampled span for container log collectors; OTLP spans preserve runtime timing, carry explicit AnyCode correlation attributes, and expose flush/shutdown lifecycle controls. Configurable span/event/series/histogram retention prevents long-lived telemetry growth and exposes drop counters, while exporter failures remain isolated from run behavior.
+- **End-to-end cancellation ownership** — caller cancellation now propagates through `AgentRunner`, high-level run and stream APIs, orchestrator waves, provider waits, parallel tools, and shell process trees without being converted into an ordinary result. Agents settle in an explicit `cancelled` state, durable runs persist a `user_cancelled` stop and checkpoint, semaphore accounting remains balanced, and `AnyCode.close()` cancels and awaits all tracked standalone, coordinator, team, reflection, and handoff operations before resource teardown.
+- **Fail-closed side-effect idempotency** — tools can opt into atomic claim-before-execute semantics with `side_effecting=True`. Explicit business keys take precedence over deterministic run/turn/call fallbacks; completed calls replay, mismatched input conflicts, and in-progress or unrecorded outcomes terminate the run with `side_effect_unknown` instead of being retried. Post-invocation errors default to non-retryable unless the tool explicitly proves otherwise, and uncertain claims are retained during pruning for operator reconciliation. Public in-memory and SQLite stores support process-local or restart-safe coordination, hashed storage keys, redacted result persistence, and pruning. Mutating built-ins and every discovered MCP tool are protected by default.
+- **Provider capacity controls** — `ResilientAdapter` now applies a provider-scoped concurrency bulkhead (default `8`) and optional evenly paced `requests_per_minute` limit to chat and streaming attempts, including retries. Capacity is shared across adapters in the same event loop and scope; conflicting limits for one scope fail clearly. Queue waits load-shed with `ProviderCapacityError` after a configurable timeout, while cancellation and early stream closure release slots immediately. `ProviderResilienceConfig` is configurable globally through `OrchestratorConfig`/YAML or per `AgentConfig`, with per-agent precedence.
+- **Protected and bounded durable storage** — `AgentRunner` and `RunScheduler` now accept the public `RunStore` protocol so production backends can replace the local filesystem implementation. `FilesystemRunStore` accepts a `RunPayloadProtector` for versioned, fail-closed protected payload envelopes while retaining read compatibility with legacy plaintext stores. Run records, transcript events, and turn checkpoints now carry an explicit schema version; legacy unversioned artifacts remain readable and unsupported future versions fail clearly. Workflow checkpoints now default correctly to format v2 while retaining v1 compatibility. `RunRetentionPolicy` prunes only terminal runs by age and count, and can be applied through `sweep_once`, `RunScheduler`, or `anycode runs sweep`. Run IDs are constrained to one path segment to prevent storage-root traversal.
+- **Default-on credential redaction** — centralized `redact_text`, `redact_sensitive`, and `safe_exception_message` helpers now protect built-in telemetry exports, exception surfaces, workflow and turn checkpoints, run records and transcripts, context artifacts, session-chain files, persistent memory backends, eval reports, and harness artifacts. Structured sensitive keys and common provider/cloud token formats are replaced with `<redacted-secret>` while token-usage metrics and other non-secret fields retain their shape. Persistence and exporter configs expose explicit `redact_sensitive_data=False` opt-outs for independently protected stores that require exact replay.
+
 ## [0.6.0] - 2026-07-10
 
 ### Added
@@ -33,6 +47,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- Declarative config loading now rejects unknown root, agent, task, context-engineering, and nested model fields with `UnknownConfigFieldError` instead of silently ignoring them. Programmatic model construction is unchanged.
 - **Handoff sentinel encoding** — the built-in `handoff` tool now encodes its payload as `__HANDOFF__:<json>` instead of a colon-delimited string. Free-form `summary`/`reason` text containing `:` (or any other character) now round-trips losslessly through `AgentRunner._detect_handoff`. New helpers `encode_handoff_payload`/`decode_handoff_payload` are exported from `anycode.handoff.tool`.
 - **`AgentState`** uses `Field(default_factory=...)` for `messages` and `token_usage` to guarantee per-instance defaults.
 - **`AgentRunner.stream` handoff path** now appends every executed `ToolCallRecord` from the same batch and emits a `tool_result` event for each before yielding the `handoff`/`done` events. Previously, sibling tool calls executed alongside a handoff were dropped from `RunResult.tool_calls` and never streamed to observers.
@@ -196,7 +211,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Four examples: solo worker, crew workflow, staged pipeline, hybrid tooling.
 - Pydantic-based immutable type system (`frozen=True` on all models).
 
-[Unreleased]: https://github.com/Quantlix/anycode/compare/v0.6.0...HEAD
+[Unreleased]: https://github.com/Quantlix/anycode/compare/v0.7.0...HEAD
+[0.7.0]: https://github.com/Quantlix/anycode/compare/v0.6.0...v0.7.0
 [0.6.0]: https://github.com/Quantlix/anycode/compare/v0.5.0...v0.6.0
 [0.5.0]: https://github.com/Quantlix/anycode/compare/v0.4.0...v0.5.0
 [0.4.0]: https://github.com/Quantlix/anycode/compare/v0.3.0...v0.4.0

{anycode_py-0.6.0 → anycode_py-0.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: anycode-py
-Version: 0.6.0
+Version: 0.7.0
 Summary: Orchestrate autonomous AI agent teams with dependency-aware task scheduling, inter-agent messaging, and provider-agnostic LLM integration.
 Project-URL: Homepage, https://github.com/Quantlix/anycode
 Project-URL: Repository, https://github.com/Quantlix/anycode
@@ -18,10 +18,27 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Software Development :: Libraries
 Classifier: Typing :: Typed
 Requires-Python: >=3.12
-Requires-Dist: anthropic>=0.40
-Requires-Dist: openai>=1.50
 Requires-Dist: pydantic>=2.0
 Requires-Dist: python-dotenv>=1.2.2
+Provides-Extra: all
+Requires-Dist: aiosqlite>=0.20; extra == 'all'
+Requires-Dist: anthropic>=0.40; extra == 'all'
+Requires-Dist: boto3>=1.34; extra == 'all'
+Requires-Dist: chromadb>=0.5; extra == 'all'
+Requires-Dist: google-genai>=1.0; extra == 'all'
+Requires-Dist: httpx>=0.27; extra == 'all'
+Requires-Dist: mcp>=1.0; extra == 'all'
+Requires-Dist: openai>=1.50; extra == 'all'
+Requires-Dist: opentelemetry-api>=1.20; extra == 'all'
+Requires-Dist: opentelemetry-exporter-otlp-proto-grpc>=1.20; extra == 'all'
+Requires-Dist: opentelemetry-sdk>=1.20; extra == 'all'
+Requires-Dist: pyyaml>=6.0; extra == 'all'
+Requires-Dist: redis[hiredis]>=5.0; extra == 'all'
+Requires-Dist: rich>=13.0; extra == 'all'
+Requires-Dist: tiktoken>=0.8; extra == 'all'
+Requires-Dist: typer>=0.12; extra == 'all'
+Provides-Extra: anthropic
+Requires-Dist: anthropic>=0.40; extra == 'anthropic'
 Provides-Extra: azure
 Requires-Dist: openai>=1.50; extra == 'azure'
 Provides-Extra: bedrock
@@ -36,8 +53,16 @@ Provides-Extra: mcp
 Requires-Dist: mcp>=1.0; extra == 'mcp'
 Provides-Extra: ollama
 Requires-Dist: httpx>=0.27; extra == 'ollama'
+Provides-Extra: openai
+Requires-Dist: openai>=1.50; extra == 'openai'
 Provides-Extra: persistence
 Requires-Dist: aiosqlite>=0.20; extra == 'persistence'
+Provides-Extra: providers
+Requires-Dist: anthropic>=0.40; extra == 'providers'
+Requires-Dist: boto3>=1.34; extra == 'providers'
+Requires-Dist: google-genai>=1.0; extra == 'providers'
+Requires-Dist: httpx>=0.27; extra == 'providers'
+Requires-Dist: openai>=1.50; extra == 'providers'
 Provides-Extra: redis
 Requires-Dist: redis[hiredis]>=5.0; extra == 'redis'
 Provides-Extra: telemetry
@@ -64,7 +89,7 @@ Description-Content-Type: text/markdown
 
 > Developed and maintained by [Quantlix](https://github.com/Quantlix).
 
-> **Active development warning:** AnyCode is an alpha-stage framework under active development. APIs, defaults, runtime behavior, and configuration formats may change between releases. It is not meant to be used for production systems yet, especially systems that handle sensitive data, irreversible actions, customer workloads, or critical infrastructure.
+> **Alpha status and production boundary:** AnyCode is under active development. Supported top-level APIs and persisted formats have explicit compatibility contracts, but pre-1.0 minor releases may require migration. Production use is workload-specific: bounded deployments are eligible only after the [production readiness checklist](https://quantlix.github.io/anycode/latest/guides/production-readiness/) passes and the operator supplies host and network isolation, identity, durable storage, secrets management, monitoring, and incident controls. Direct safety-critical, critical-infrastructure, or unrestricted irreversible use is a no-go.
 
 AnyCode is a Python framework for building coordinated AI agent teams. It helps developers compose autonomous LLM agents, connect them to typed tools, schedule dependent tasks, share memory, stream output, route work across providers, and inspect long-running agent workflows with explicit lifecycle and verification data.
 
@@ -82,7 +107,7 @@ The framework focuses on practical harness engineering:
 - Pydantic-validated tool calls and immutable runtime models.
 - Observability, guardrails, structured output, checkpointing, HITL approval, MCP tools, routing, cost tracking, RAG memory, and verification gates.
 
-AnyCode is built for experimentation, evaluation, local development, research prototypes, and early-stage product exploration. It is not yet a stable production platform.
+AnyCode is built for experimentation, evaluation, local development, research prototypes, and bounded automation. A pinned deployment can be eligible for production when its workload-specific controls and evidence pass the readiness review; the package alone is not a production guarantee.
 
 ## Current Package
 
@@ -90,7 +115,7 @@ AnyCode is built for experimentation, evaluation, local development, research pr
 | --- | --- |
 | Distribution | `anycode-py` |
 | Import package | `anycode` |
-| Current version | `0.6.0` |
+| Current version | `0.7.0` |
 | Python | `>=3.12` |
 | Project status | Alpha |
 | License | MIT |
@@ -162,7 +187,7 @@ The main design rule is simple: the framework owns the harness, while providers
 For a new or existing Python project:
 
 ```bash
-uv add anycode-py
+uv add "anycode-py[anthropic]"
 ```
 
 For CLI and YAML/TOML configuration support:
@@ -174,7 +199,7 @@ uv add "anycode-py[cli]"
 For the full optional ecosystem:
 
 ```bash
-uv add "anycode-py[cli,telemetry,persistence,redis,vector,mcp,google,ollama,bedrock,azure]"
+uv add "anycode-py[all]"
 ```
 
 ### 3. Add API Keys
@@ -437,7 +462,7 @@ Custom tools use Pydantic input models and are registered through `define_tool()
 
 ## Examples Cookbook
 
-The `examples/` directory contains 26 runnable scripts. They are arranged from beginner workflows to runtime reliability demos.
+The `examples/` directory contains 34 runnable scripts. They are arranged from beginner workflows to runtime reliability demos.
 
 | Examples | Theme |
 | --- | --- |
@@ -450,6 +475,8 @@ The `examples/` directory contains 26 runnable scripts. They are arranged from b
 | `22_deterministic_eval.py` to `25_runtime_cancellation.py` | Fake adapters, context pressure, verification gates, and cancellation telemetry |
 | `26_context_engineering.py` | Huge-context model profiles, section budgets, first-class section inputs, and usage reports |
 | `27_plugin_ecosystem.py` | Plugin bundles wiring custom tools, provider factories, and sensors into the engine |
+| `28_durable_runs.py` to `30_scheduled_wakeups.py` | Durable resumable runs, session chaining, and scheduled wakeups |
+| `31_streaming_runtime.py` to `34_list_files.py` | Provider-token streaming, reasoning-model controls, authenticated MCP over HTTP, and fast file listing |
 
 Run an example from the repository root:
 
@@ -484,20 +511,21 @@ The default pytest configuration excludes integration tests. Integration tests m
 
 AnyCode agents can be connected to tools that read files, write files, execute commands, call providers, and access external systems through MCP. Treat every tool-enabled agent as a privileged automation process.
 
-Recommended safeguards while the framework is alpha:
+Required safeguards for any deployment:
 
-- Run agents in disposable workspaces.
-- Scope tool access to the minimum needed tools.
+- Run tool-enabled agents as a non-root identity inside a container, VM, or equivalent isolation boundary.
+- Set explicit agent tool lists and apply `ToolSecurityPolicy` with narrow path, shell, and environment access.
 - Use fake adapters or deterministic evaluation before live model calls.
-- Keep API keys in environment variables or `.env` files that are not committed.
-- Add human approval gates for sensitive or irreversible actions.
-- Avoid production workloads until the security roadmap has landed and been independently reviewed.
+- Inject API keys from a protected secret source and keep them out of prompts, source, images, and logs.
+- Use durable idempotency plus human approval for sensitive or irreversible actions.
+- Allowlist production plugins and MCP endpoints, then enforce network egress outside the framework.
+- Review the [security and threat model](https://quantlix.github.io/anycode/latest/reference/security/) and complete the [production readiness checklist](https://quantlix.github.io/anycode/latest/guides/production-readiness/).
 
 ## FAQ
 
 ### Is AnyCode production ready?
 
-No. AnyCode is under active development and currently marked alpha. It includes production-oriented ideas such as telemetry, guardrails, checkpoints, approvals, and quality gates, but the project is not meant for production use yet.
+Production readiness is workload-specific. AnyCode remains alpha, but a pinned release can support bounded, reversible, operator-monitored workloads when every mandatory readiness control passes. Customer-facing, multi-tenant, sensitive-data, or irreversible workflows require stronger application-owned isolation, authorization, storage, and operations. Direct safety-critical control, critical infrastructure control, and autonomous high-impact decisions are no-go uses for AnyCode as the sole decision or control system.
 
 ### What makes AnyCode different from a single-agent loop?
 

{anycode_py-0.6.0 → anycode_py-0.7.0}/README.md

@@ -12,7 +12,7 @@
 
 > Developed and maintained by [Quantlix](https://github.com/Quantlix).
 
-> **Active development warning:** AnyCode is an alpha-stage framework under active development. APIs, defaults, runtime behavior, and configuration formats may change between releases. It is not meant to be used for production systems yet, especially systems that handle sensitive data, irreversible actions, customer workloads, or critical infrastructure.
+> **Alpha status and production boundary:** AnyCode is under active development. Supported top-level APIs and persisted formats have explicit compatibility contracts, but pre-1.0 minor releases may require migration. Production use is workload-specific: bounded deployments are eligible only after the [production readiness checklist](https://quantlix.github.io/anycode/latest/guides/production-readiness/) passes and the operator supplies host and network isolation, identity, durable storage, secrets management, monitoring, and incident controls. Direct safety-critical, critical-infrastructure, or unrestricted irreversible use is a no-go.
 
 AnyCode is a Python framework for building coordinated AI agent teams. It helps developers compose autonomous LLM agents, connect them to typed tools, schedule dependent tasks, share memory, stream output, route work across providers, and inspect long-running agent workflows with explicit lifecycle and verification data.
 
@@ -30,7 +30,7 @@ The framework focuses on practical harness engineering:
 - Pydantic-validated tool calls and immutable runtime models.
 - Observability, guardrails, structured output, checkpointing, HITL approval, MCP tools, routing, cost tracking, RAG memory, and verification gates.
 
-AnyCode is built for experimentation, evaluation, local development, research prototypes, and early-stage product exploration. It is not yet a stable production platform.
+AnyCode is built for experimentation, evaluation, local development, research prototypes, and bounded automation. A pinned deployment can be eligible for production when its workload-specific controls and evidence pass the readiness review; the package alone is not a production guarantee.
 
 ## Current Package
 
@@ -38,7 +38,7 @@ AnyCode is built for experimentation, evaluation, local development, research pr
 | --- | --- |
 | Distribution | `anycode-py` |
 | Import package | `anycode` |
-| Current version | `0.6.0` |
+| Current version | `0.7.0` |
 | Python | `>=3.12` |
 | Project status | Alpha |
 | License | MIT |
@@ -110,7 +110,7 @@ The main design rule is simple: the framework owns the harness, while providers
 For a new or existing Python project:
 
 ```bash
-uv add anycode-py
+uv add "anycode-py[anthropic]"
 ```
 
 For CLI and YAML/TOML configuration support:
@@ -122,7 +122,7 @@ uv add "anycode-py[cli]"
 For the full optional ecosystem:
 
 ```bash
-uv add "anycode-py[cli,telemetry,persistence,redis,vector,mcp,google,ollama,bedrock,azure]"
+uv add "anycode-py[all]"
 ```
 
 ### 3. Add API Keys
@@ -385,7 +385,7 @@ Custom tools use Pydantic input models and are registered through `define_tool()
 
 ## Examples Cookbook
 
-The `examples/` directory contains 26 runnable scripts. They are arranged from beginner workflows to runtime reliability demos.
+The `examples/` directory contains 34 runnable scripts. They are arranged from beginner workflows to runtime reliability demos.
 
 | Examples | Theme |
 | --- | --- |
@@ -398,6 +398,8 @@ The `examples/` directory contains 26 runnable scripts. They are arranged from b
 | `22_deterministic_eval.py` to `25_runtime_cancellation.py` | Fake adapters, context pressure, verification gates, and cancellation telemetry |
 | `26_context_engineering.py` | Huge-context model profiles, section budgets, first-class section inputs, and usage reports |
 | `27_plugin_ecosystem.py` | Plugin bundles wiring custom tools, provider factories, and sensors into the engine |
+| `28_durable_runs.py` to `30_scheduled_wakeups.py` | Durable resumable runs, session chaining, and scheduled wakeups |
+| `31_streaming_runtime.py` to `34_list_files.py` | Provider-token streaming, reasoning-model controls, authenticated MCP over HTTP, and fast file listing |
 
 Run an example from the repository root:
 
@@ -432,20 +434,21 @@ The default pytest configuration excludes integration tests. Integration tests m
 
 AnyCode agents can be connected to tools that read files, write files, execute commands, call providers, and access external systems through MCP. Treat every tool-enabled agent as a privileged automation process.
 
-Recommended safeguards while the framework is alpha:
+Required safeguards for any deployment:
 
-- Run agents in disposable workspaces.
-- Scope tool access to the minimum needed tools.
+- Run tool-enabled agents as a non-root identity inside a container, VM, or equivalent isolation boundary.
+- Set explicit agent tool lists and apply `ToolSecurityPolicy` with narrow path, shell, and environment access.
 - Use fake adapters or deterministic evaluation before live model calls.
-- Keep API keys in environment variables or `.env` files that are not committed.
-- Add human approval gates for sensitive or irreversible actions.
-- Avoid production workloads until the security roadmap has landed and been independently reviewed.
+- Inject API keys from a protected secret source and keep them out of prompts, source, images, and logs.
+- Use durable idempotency plus human approval for sensitive or irreversible actions.
+- Allowlist production plugins and MCP endpoints, then enforce network egress outside the framework.
+- Review the [security and threat model](https://quantlix.github.io/anycode/latest/reference/security/) and complete the [production readiness checklist](https://quantlix.github.io/anycode/latest/guides/production-readiness/).
 
 ## FAQ
 
 ### Is AnyCode production ready?
 
-No. AnyCode is under active development and currently marked alpha. It includes production-oriented ideas such as telemetry, guardrails, checkpoints, approvals, and quality gates, but the project is not meant for production use yet.
+Production readiness is workload-specific. AnyCode remains alpha, but a pinned release can support bounded, reversible, operator-monitored workloads when every mandatory readiness control passes. Customer-facing, multi-tenant, sensitive-data, or irreversible workflows require stronger application-owned isolation, authorization, storage, and operations. Direct safety-critical control, critical infrastructure control, and autonomous high-impact decisions are no-go uses for AnyCode as the sole decision or control system.
 
 ### What makes AnyCode different from a single-agent loop?
 

{anycode_py-0.6.0 → anycode_py-0.7.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "anycode-py"
-version = "0.6.0"
+version = "0.7.0"
 description = "Orchestrate autonomous AI agent teams with dependency-aware task scheduling, inter-agent messaging, and provider-agnostic LLM integration."
 readme = "README.md"
 license = "MIT"
@@ -33,8 +33,6 @@ classifiers = [
 ]
 dependencies = [
     "pydantic>=2.0",
-    "anthropic>=0.40",
-    "openai>=1.50",
     "python-dotenv>=1.2.2",
 ]
 
@@ -48,6 +46,12 @@ Changelog = "https://github.com/Quantlix/anycode/blob/main/CHANGELOG.md"
 anycode = "anycode.cli.main:app"
 
 [project.optional-dependencies]
+anthropic = [
+    "anthropic>=0.40",
+]
+openai = [
+    "openai>=1.50",
+]
 telemetry = [
     "opentelemetry-api>=1.20",
     "opentelemetry-exporter-otlp-proto-grpc>=1.20",
@@ -77,6 +81,13 @@ bedrock = [
 azure = [
     "openai>=1.50",
 ]
+providers = [
+    "anthropic>=0.40",
+    "openai>=1.50",
+    "google-genai>=1.0",
+    "httpx>=0.27",
+    "boto3>=1.34",
+]
 mcp = [
     "mcp>=1.0",
 ]
@@ -85,9 +96,29 @@ cli = [
     "rich>=13.0",
     "pyyaml>=6.0",
 ]
+all = [
+    "anthropic>=0.40",
+    "openai>=1.50",
+    "google-genai>=1.0",
+    "httpx>=0.27",
+    "boto3>=1.34",
+    "opentelemetry-api>=1.20",
+    "opentelemetry-exporter-otlp-proto-grpc>=1.20",
+    "opentelemetry-sdk>=1.20",
+    "aiosqlite>=0.20",
+    "redis[hiredis]>=5.0",
+    "tiktoken>=0.8",
+    "chromadb>=0.5",
+    "mcp>=1.0",
+    "typer>=0.12",
+    "rich>=13.0",
+    "pyyaml>=6.0",
+]
 
 [dependency-groups]
 dev = [
+    "anthropic>=0.40",
+    "openai>=1.50",
     "pyright>=1.1.408",
     "pytest>=8.0",
     "pytest-asyncio>=0.24",
@@ -105,6 +136,10 @@ dev = [
     "build>=1.4.2",
     "hatchling>=1.29.0",
     "twine>=6.2.0",
+    "mkdocs>=1.6.1",
+    "mkdocs-material[imaging]>=9.7.6",
+    "mkdocstrings[python]>=1.0.4",
+    "mike>=2.1.3",
 ]
 
 [tool.hatch.build.targets.wheel]

{anycode_py-0.6.0 → anycode_py-0.7.0}/src/anycode/__init__.py

@@ -1,11 +1,13 @@
 """AnyCode — scalable multi-agent AI orchestration framework for Python."""
 
 from anycode.checkpoint.manager import CheckpointManager
+from anycode.checkpoint.serializer import UnsupportedCheckpointVersionError
 from anycode.checkpoint.store import FilesystemCheckpointStore
 from anycode.collaboration.kv_store import InMemoryStore
 from anycode.collaboration.message_bus import MessageBus
 from anycode.collaboration.shared_mem import SharedMemory
 from anycode.collaboration.team import Team
+from anycode.config.loader import UnknownConfigFieldError, UnsupportedConfigVersionError
 from anycode.constants import (
     ORCH_EVENT_AGENT_COMPLETE,
     ORCH_EVENT_AGENT_START,
@@ -106,16 +108,24 @@ from anycode.plugins import (
 )
 from anycode.providers.adapter import create_adapter
 from anycode.providers.fake import FakeAdapter, FakeResponse
-from anycode.providers.resilience import ProviderUnavailableError, ResilientAdapter
+from anycode.providers.resilience import (
+    ProviderCapacityConfigurationError,
+    ProviderCapacityError,
+    ProviderCapacityLimiter,
+    ProviderUnavailableError,
+    ResilientAdapter,
+)
 from anycode.reflection.critic import DEFAULT_CRITIC_PROMPT, LLMCritic
 from anycode.reflection.evaluator import parse_critic_json
 from anycode.reflection.loop import ReflectionLoop
 from anycode.routing.classifier import classify_task
 from anycode.routing.router import DefaultRouter
 from anycode.routing.rules import evaluate_rules, match_rule
-from anycode.runstore.store import FilesystemRunStore
+from anycode.runstore.protocol import RunPayloadProtector, RunStore
+from anycode.runstore.store import FilesystemRunStore, ProtectedPayloadError, UnsupportedRunStoreVersionError
 from anycode.schedule.scheduler import RunScheduler, SweepReport, sweep_once
 from anycode.schedule.tasks import ScheduledTask, ScheduledTaskResult, run_scheduled_task
+from anycode.security import REDACTED_SECRET, redact_sensitive, redact_text, safe_exception_message
 from anycode.structured.output import (
     STRUCTURED_OUTPUT_TOOL_NAME,
     build_retry_prompt,
@@ -127,9 +137,16 @@ from anycode.tasks.queue import TaskQueue
 from anycode.tasks.task import create_task, get_task_dependency_order, is_task_ready, validate_task_dependencies
 from anycode.telemetry.events import EventEmitter, TelemetryEvent
 from anycode.telemetry.metrics import MetricsCollector, Timer
-from anycode.telemetry.tracer import ConsoleExporter, Span, Tracer
+from anycode.telemetry.tracer import ConsoleExporter, JSONLExporter, OTLPExporter, Span, Tracer
 from anycode.tools.built_in import BUILT_IN_TOOLS, register_built_in_tools
 from anycode.tools.executor import ToolExecutor
+from anycode.tools.idempotency import (
+    IdempotencyClaim,
+    InMemoryToolIdempotencyStore,
+    SQLiteToolIdempotencyStore,
+    ToolIdempotencyStore,
+    create_tool_idempotency_store,
+)
 from anycode.tools.registry import ToolRegistry, define_tool
 from anycode.types import (
     AcceptanceThresholds,
@@ -201,6 +218,7 @@ from anycode.types import (
     LLMToolDef,
     MCPServerConfig,
     MCPToolInfo,
+    MCPTrustPolicy,
     MemoryConfig,
     MemoryEntry,
     MemoryStore,
@@ -214,6 +232,7 @@ from anycode.types import (
     PluginInstallation,
     PluginManifest,
     PluginSource,
+    PluginTrustPolicy,
     PoolStatus,
     ProviderFactory,
     ProviderResilienceConfig,
@@ -221,6 +240,7 @@ from anycode.types import (
     RAGConfig,
     RAGContext,
     RAGEntry,
+    RedactedThinkingBlock,
     ReflectionConfig,
     RetryPolicy,
     RouteDecision,
@@ -228,8 +248,10 @@ from anycode.types import (
     RoutingConfig,
     RoutingRule,
     RunnerOptions,
+    RunnerStreamingConfig,
     RunRecord,
     RunResult,
+    RunRetentionPolicy,
     RunStatus,
     RunSummary,
     SchedulingStrategy,
@@ -247,13 +269,17 @@ from anycode.types import (
     TaskStatus,
     TeamConfig,
     TeamRunResult,
+    TerminalRunStatus,
     TextBlock,
+    ThinkingBlock,
     TokenizerStrategy,
     TokenUsage,
     ToolCallRecord,
     ToolDefinition,
+    ToolIdempotencyConfig,
     ToolResult,
     ToolResultBlock,
+    ToolSecurityPolicy,
     ToolUseBlock,
     ToolUseContext,
     TraceConfig,
@@ -321,6 +347,11 @@ __all__ = [
     "rebuild_from_handoff",
     "offload_text",
     "restore_text",
+    # Data protection
+    "REDACTED_SECRET",
+    "redact_sensitive",
+    "redact_text",
+    "safe_exception_message",
     # Verification & quality gates
     "QualityGate",
     "Sensor",
@@ -357,17 +388,29 @@ __all__ = [
     "FakeResponse",
     # Provider resilience
     "ResilientAdapter",
+    "ProviderCapacityConfigurationError",
+    "ProviderCapacityError",
+    "ProviderCapacityLimiter",
     "ProviderUnavailableError",
     "ProviderResilienceConfig",
     "RetryPolicy",
     # Durable run store
     "FilesystemRunStore",
+    "ProtectedPayloadError",
+    "RunPayloadProtector",
+    "RunStore",
     "DurabilityConfig",
     "RunRecord",
+    "RunRetentionPolicy",
     "RunStatus",
+    "TerminalRunStatus",
     "TranscriptEvent",
     "TurnCheckpoint",
     "BudgetSnapshot",
+    "UnsupportedCheckpointVersionError",
+    "UnsupportedConfigVersionError",
+    "UnsupportedRunStoreVersionError",
+    "UnknownConfigFieldError",
     # Session chaining
     "SessionChain",
     "GoalContract",
@@ -415,7 +458,6 @@ __all__ = [
     # Memory
     "CompositeMemory",
     "InMemoryVectorStore",
-    "create_memory_store",
     # Checkpoint
     "CheckpointManager",
     "FilesystemCheckpointStore",
@@ -433,6 +475,11 @@ __all__ = [
     # Tools
     "ToolRegistry",
     "ToolExecutor",
+    "IdempotencyClaim",
+    "InMemoryToolIdempotencyStore",
+    "SQLiteToolIdempotencyStore",
+    "ToolIdempotencyStore",
+    "create_tool_idempotency_store",
     "define_tool",
     "BUILT_IN_TOOLS",
     "register_built_in_tools",
@@ -445,6 +492,8 @@ __all__ = [
     "Tracer",
     "Span",
     "ConsoleExporter",
+    "JSONLExporter",
+    "OTLPExporter",
     "MetricsCollector",
     "Timer",
     "EventEmitter",
@@ -470,6 +519,8 @@ __all__ = [
     "ToolUseBlock",
     "ToolResultBlock",
     "ImageBlock",
+    "ThinkingBlock",
+    "RedactedThinkingBlock",
     "LLMMessage",
     "TokenUsage",
     "LLMResponse",
@@ -478,7 +529,9 @@ __all__ = [
     "ToolUseContext",
     "AgentInfo",
     "ToolResult",
+    "ToolSecurityPolicy",
     "ToolDefinition",
+    "ToolIdempotencyConfig",
     "AgentConfig",
     "AgentState",
     "ToolCallRecord",
@@ -505,6 +558,7 @@ __all__ = [
     "LLMStreamOptions",
     "LLMAdapter",
     "MCPServerConfig",
+    "MCPTrustPolicy",
     "MCPToolInfo",
     "HandoffRequest",
     "Handoff",
@@ -516,6 +570,7 @@ __all__ = [
     "PluginManifest",
     "PluginRegistry",
     "PluginSource",
+    "PluginTrustPolicy",
     "ProviderFactory",
     "discover_entry_point_plugins",
     "get_provider_factory",
@@ -527,6 +582,7 @@ __all__ = [
     "RouteDecision",
     "Router",
     "RunnerOptions",
+    "RunnerStreamingConfig",
     "RunResult",
     "PoolStatus",
     "SchedulingStrategy",

{anycode_py-0.6.0 → anycode_py-0.7.0}/src/anycode/checkpoint/__init__.py

@@ -1,11 +1,13 @@
 """Workflow checkpointing and resume for crash recovery."""
 
 from anycode.checkpoint.manager import CheckpointManager
+from anycode.checkpoint.serializer import UnsupportedCheckpointVersionError
 from anycode.checkpoint.store import FilesystemCheckpointStore
 
 __all__ = [
     "CheckpointManager",
     "FilesystemCheckpointStore",
+    "UnsupportedCheckpointVersionError",
 ]
 
 try:

{anycode_py-0.6.0 → anycode_py-0.7.0}/src/anycode/checkpoint/manager.py

@@ -17,7 +17,10 @@ class CheckpointManager:
 
     def __init__(self, config: CheckpointConfig, store: CheckpointStore | None = None) -> None:
         self._config = config
-        self._store: CheckpointStore = store or FilesystemCheckpointStore(config.path)
+        self._store: CheckpointStore = store or FilesystemCheckpointStore(
+            config.path,
+            redact_sensitive_data=config.redact_sensitive_data,
+        )
 
     @property
     def store(self) -> CheckpointStore: