Python SDK 0.4.0

What's Changed

• Create AGENTS.md by @PetrefiedThunder in #107
• Add scoped AGENTS.md (app/services + tests) by @PetrefiedThunder in #108
• Align AWI control plane with research draft by @PetrefiedThunder in #109
• Refresh trust plane README by @PetrefiedThunder in #110
• Sharpen the trust-plane product spine by @PetrefiedThunder in #112
• fix(security): require agent ownership on comms send/ack/handoff by @PetrefiedThunder in #114
• build(deps): update pydantic-settings requirement from >=2.14.0 to >=2.14.1 by @dependabot[bot] in #116
• build(deps): update cryptography requirement from >=46.0.0 to >=48.0.0 by @dependabot[bot] in #117
• build(deps): update uvicorn requirement from >=0.46.0 to >=0.48.0 by @dependabot[bot] in #118
• build(deps): update greenlet requirement from >=3.0.0 to >=3.5.1 by @dependabot[bot] in #119
• build(deps): update fastapi requirement from >=0.136.1 to >=0.136.3 by @dependabot[bot] in #120
• Extract shared wrapper edge client and document KMS path by @PetrefiedThunder in #128
• Bump deps to clear 18 of 19 known CVEs by @PetrefiedThunder in #137
• Flip trust-plane defaults to fail-closed by @PetrefiedThunder in #144
• build(deps): bump pyjwt from 2.12.1 to 2.13.0 by @dependabot[bot] in #133
• build(deps): bump starlette from 1.0.0 to 1.3.1 by @dependabot[bot] in #136
• Bump cryptography & python-multipart in uv.lock (clear CVEs) by @PetrefiedThunder in #145
• build(deps): update redis requirement from >=7.4.0 to >=8.0.0 by @dependabot[bot] in #121
• build(deps): update pytest-asyncio requirement from >=1.3.0 to >=1.4.0 by @dependabot[bot] in #124
• build(deps): update uvicorn requirement from >=0.48.0 to >=0.49.0 by @dependabot[bot] in #127
• build(deps): update stripe requirement from >=15.1.0 to >=15.2.1 by @dependabot[bot] in #129
• build(deps): update greenlet requirement from >=3.5.1 to >=3.5.2 by @dependabot[bot] in #140
• build(deps): bump pydantic-settings from 2.14.1 to 2.14.2 by @dependabot[bot] in #138
• build(deps): update fastapi requirement from >=0.136.3 to >=0.138.1 by @dependabot[bot] in #139
• build(deps): update pydantic-settings requirement from >=2.14.1 to >=2.14.2 by @dependabot[bot] in #142
• build(deps): bump actions/checkout from 6 to 7 by @dependabot[bot] in #141
• build(deps): update pytest requirement from >=9.0.3 to >=9.1.1 by @dependabot[bot] in #143
• Dev-loop efficiency: fast test split, fix mypy hook, quiet warnings by @PetrefiedThunder in #146
• build(deps): bump actions/cache from 5 to 6 by @dependabot[bot] in #147
• build(deps): update sqlmodel requirement from >=0.0.38 to >=0.0.39 by @dependabot[bot] in #148
• build(deps): update stripe requirement from >=15.2.1 to >=15.3.0 by @dependabot[bot] in #149
• build(deps): update redis requirement from >=8.0.0 to >=8.0.1 by @dependabot[bot] in #150
• Phase 1 dev-loop: planner solver fallback + self-contained make test by @PetrefiedThunder in #153
• build(deps): update alembic requirement from >=1.18.4 to >=1.18.5 by @dependabot[bot] in #151
• build(deps): update greenlet requirement from >=3.5.1 to >=3.5.3 by @dependabot[bot] in #152
• Phase 2 (#3): single dependency source — stop committing uv.lock by @PetrefiedThunder in #154
• Repo review + fix KYC IDOR, billing idempotency, and enable real lint by @PetrefiedThunder in #157
• build(deps): update fastapi requirement from >=0.138.1 to >=0.139.0 by @dependabot[bot] in #155
• Remove dead proof-surface code and re-harden mypy by @PetrefiedThunder in #158
• build(deps): update mypy requirement from >=2.1.0 to >=2.2.0 by @dependabot[bot] in #160
• build(deps): update uvicorn requirement from >=0.49.0 to >=0.51.0 by @dependabot[bot] in #161
• Trust/billing/auth hardening: tenant isolation, idempotency, concurrency, audit-chain, and carry #159 to main by @PetrefiedThunder in #162
• build(deps): update stripe requirement from >=15.3.0 to >=15.3.1 by @dependabot[bot] in #163
• build(deps): update mypy requirement from >=2.2.0 to >=2.3.0 by @dependabot[bot] in #165
• build(deps): update uvicorn requirement from >=0.49.0 to >=0.51.0 by @dependabot[bot] in #166
• build(deps): bump actions/setup-python from 6 to 7 by @dependabot[bot] in #167
• build(deps): update fastapi requirement from >=0.139.0 to >=0.141.1 by @dependabot[bot] in #168
• build(deps): update greenlet requirement from >=3.5.3 to >=3.5.4 by @dependabot[bot] in #169
• Harden trust-plane scaffolding and close AWI HTTP bypass by @PetrefiedThunder in #170
• docs: partner first-tool runbook by @PetrefiedThunder in #171
• docs: link partner runbook from WEDGE by @PetrefiedThunder in #172
• Add dogfood trust-plane playground (partner.notes.write) by @PetrefiedThunder in #173
• Harden trust-plane charge/receipt and idempotency (P0) by @PetrefiedThunder in #174
• Sync exactly-once positioning + marketing site draft by @PetrefiedThunder in #175
• Publish site agent discovery pointers by @PetrefiedThunder in #176
• docs(site): agent-middleware naming and Vercel URL by @PetrefiedThunder in #177
• Normalize durable-state DATABASE_URL and fail closed by @PetrefiedThunder in #178
• docs: Phase 1 live verify + Phase 2 partner inventory caveats by @PetrefiedThunder in #179
• fix: gate MCP discovery stubs when proof surfaces are off by @PetrefiedThunder in #180
• docs: Phase 2 live acceptance verified on Railway by @PetrefiedThunder in #181
• Phase 3: Railway deploy posture + production-trust CI by @PetrefiedThunder in #186
• Phase 4: migrations-on-start / reduce create_all reliance by @PetrefiedThunder in #187
• Phase 5: docs / OpenAPI / discovery wedge honesty by @PetrefiedThunder in #188
• Phase 6: freeze proof-surface scaffolding by @PetrefiedThunder in #189
• Phase 7: optional follow-ups (safe slice) by @PetrefiedThunder in #190
• build(deps): update redis requirement from >=8.0.1 to >=8.1.0 by @dependabot[bot] in #182
• build(deps): update uvicorn requirement from >=0.51.0 to >=0.52.0 by @dependabot[bot] in https://github.com/PetrefiedThunder/agent-middle...

v1.2.0-rc.1 - Trust Plane Proof

Trust Plane Release Candidate

This prerelease packages the governed MCP trust-plane proof on main.

Verified from a clean origin/main worktree at 3bec7a9966cbd00468d2a7463a6462089190c07b.

Proof Loop

discover -> authenticate -> permit -> invoke -> meter -> receipt -> audit -> replay -> deny

Verification

• scripts/trust_release_gate.sh passed.
• Focused trust-plane suite: 55 passed.
• Trust-core coverage gate: 60 passed, 85.12% coverage over trust-plane control modules.
• Executable trust-plane demo proof passed.
• Discovery drift checks passed.
• OpenAPI parity passed.
• Simulation inventory parity passed.

Demo Proof Highlights

• Signed permit scoped to one MCP tool.
• Governed MCP invoke charged the wallet once.
• Signed receipt linked to ledger and audit artifacts.
• Replay returned the original receipt without a duplicate debit.
• Out-of-scope MCP call returned a denial receipt.
• Denial replay returned the same denial receipt.
• Cross-wallet read was denied with 403.

Caveat

This is a production-beta release candidate for the control-plane proof, not a claim of complete production settlement, full IAM replacement, or universal policy coverage.

v1.1.0

v1.1.0 — Production Hardening Release

Overview

This release focuses on production-readiness improvements including observability, resilience patterns, and API enhancements.

New Features

Structured Logging

• Added structlog for JSON-formatted structured logging
• Automatic fallback to standard logging when structlog unavailable
• Consistent log format across all services

Health Endpoints

• /health/live - Basic liveness check
• /health/ready - Readiness check with dependency verification (DB, Redis, MQTT)
• Proper health monitoring for Kubernetes/container orchestration

Resilience Patterns

• Retry with exponential backoff - Configurable retry decorator for async functions
• Circuit breaker - Protects against cascading failures in LLM service
• Automatic recovery after configurable timeout

API Enhancements

• MCP Tools pagination - New limit/offset query parameters on GET /mcp/tools
• Prevents oversized payloads on large tool registries

Graceful Shutdown

• 2-second connection drain period on shutdown
• Proper cleanup of background tasks and connections
• Structured shutdown logging

Dependencies

• Added structlog as optional dependency
• Added tenacity as optional dependency

Contributors

• Production hardening implementation

---

Full Changelog: v1.0.0...v1.1.0

v1.0.0 - Agent-Native Middleware Platform

v1.0.0 — Agent-Native Middleware Platform

Major Release

This release completes the full Agentic Web Interface (AWI) vision from arXiv:2506.10953v1.

What's New

Phase 9: AWI Complete

Feature	Description
Passkey Auth	FIDO2/WebAuthn for high-risk actions (/v1/awi/passkey/*)
DOM Bridge	Playwright-powered browser automation (/v1/awi/dom/*)
RAG Memory	ChromaDB semantic search (/v1/awi/rag/*)
Agent Discoverability	MCP tools, agent.json, llm.txt

Production Hardening

• Auth fail-safe — Rejects requests if VALID_API_KEYS unset
• Background cleanup — Periodic task cleans expired sessions
• ChromaDB persistence — Vector storage for semantic memory

9 New MCP Tools

• awi_passkey_challenge, awi_passkey_verify
• awi_dom_bridge_session, awi_dom_sync, awi_dom_state
• awi_memory_index, awi_rag_query, awi_session_context

Stats

• 339 tests passing
• Full arXiv paper alignment
• Docker-ready

Upgrade Notes

pip install -r requirements.txt

Set VALID_API_KEYS in production to enable authentication.

---

Built with arXiv:2506.10953v1 in mind.

v0.4.0 — External AWI Adoption Kit

v0.4.0 — External AWI Adoption Kit (Phase 8)

Making every website agent-native in <30 minutes.

Based on arXiv:2506.10953v1 ("Build the web for agents, not agents for the web")

What's New

• AWI Manifest Generator (python -m agent_middleware generate-awi)
• Official AWI SDK (Python + TypeScript)
• Reference templates for FastAPI and Next.js
• Zero-config adapter that translates AWI calls → our MCP + AWI backend
• Full migration guide + security checklist for website owners

Website owners can now expose a production-grade Agentic Web Interface that plugs directly into this middleware while keeping their existing human UI untouched.

Full Feature List (Cumulative)

• MCP Server Generator + @mcp_tool
• Dry-Run Sandbox + Behavioral Sandbox
• Stripe Identity KYC + API Key Rotation
• Full Agentic Web Interface (AWI) control plane
• External AWI Adoption Kit ← NEW

Tests: 317 passing

Next up: Phase 9 — Production hardening & multi-tenancy.

See the new AWI Adoption Guide to make your own site agent-ready today.

v0.2.0

v0.2.0 — Full B2A Toolchain & Economic Safety Layer

This release brings the complete end-to-end agent-native experience:

New Major Features

• MCP Server Generator (Phase 3.2/4.2)

  • @mcp_tool decorator with auto-registration
  • Unified ServiceRegistry (local SDK + persistent DB)
  • Dynamic MCP proxy (/.well-known/mcp/tools.json, /mcp/messages JSON-RPC)
  • Standalone CLI generator + FastMCP support
  • Full Pydantic → MCP JSON Schema translation
  • Seamless dry_run=True support on MCP calls

• Dry-Run Sandbox / Shadow Ledger (Phase 5.2)

  • Redis-backed stateful simulation sessions (15-min TTL)
  • async with b2a.simulate_session() as sim: context manager
  • Cumulative balance checks, cost estimates only (no side-effects)
  • Velocity isolation + single-shot /dry-run/charge endpoint

• Stripe Identity KYC (Phase 2.2)

  • Human sponsor verification before fiat top-ups
  • New wallet states + KYC_REQUIRED_FOR_TOPUP config
  • Full /v1/kyc endpoints

• API Key Rotation (Phase 5.3)

  • Automatic rotation on velocity freeze / compromise
  • Grace period + audit trail
  • SDK helpers + webhook events for agents

Other Improvements

• Unified Service Registry (replaces old register_service() logic)
• mcp optional dependency
• Expanded tests (19 new MCP tests + sandbox coverage)
• Examples + README sections for every new feature

Next up

• Phase 6 (full behavioral Sandbox Engine) + multi-tenancy scaffolding

v0.1.0 - Initial Release

v0.1.0 - Initial Release

Features

• IoT Protocol Bridge — Secure protocol translation with ACL enforcement
• Telemetry — Event ingestion, anomaly detection, auto-PR generation
• Billing — Two-tier wallets (sponsors → agents), micro-metering, arbitrage
• Agent Comms — Agent-to-agent messaging, capability discovery, swarm handoffs
• Media Engine — Video ingestion, clip extraction, distribution
• Content Factory — 1-to-20 content multiplication
• Red Team — Autonomous penetration testing
• Protocol Generator — Auto-generate OpenAPI specs from code

Technical

• 192 passing tests
• Type checking with mypy
• Linting with ruff
• Pre-commit hooks configured
• Railway deployment ready
• PostgreSQL, Redis, or memory state backend

Durability

• PostgreSQL — Recommended for production
• Redis — Caching and rate limiting
• Memory — Development/testing only