User-facing guides

Task-oriented guides for application developers building agents on the SDK, plus operator-facing docs for self-hosted / air-gapped / observability-aware setups. The full project index lives at ../README.md.

Building agents on the SDK

Read these as a sequence the first time you reach for each capability:

1. agent_skills_guide.md — packaging prompt fragments + tool bundles + defaults into Skill objects (Python or SKILL.md files); discovery; bundled resources; activation payloads.
2. mcp_guide.md — importing remote MCP tool servers over stdio, http, sse; corporate TLS / mTLS / custom-CA wiring for HTTP and SSE transports.
3. agent_middleware_guide.md — the four middleware hook points (run / stream / model / tool) and the v1.2.0 production bundle (RetryMiddleware, RateLimitMiddleware, TokenBudgetMiddleware, RedactionMiddleware).
4. human_in_the_loop_guide.md — workflow steps with pause/resume, resolver callbacks, and approval semantics.

Operating and upgrading

Read these out of order; pick what your environment needs:

• api_stability.md — what counts as the public surface; the deprecation policy. Read this once early.
• upgrading_to_1.2.md — version-specific upgrade walkthrough from 1.1.x. Read this when you bump pins.
• self_hosted_observability.md — wiring the OpenTelemetry middleware to a user-run OTLP collector, Jaeger, Grafana Tempo, or SigNoz.
• offline_install.md — air-gapped install via pip download + --no-index --find-links; no-phone-home audit.

Pages

• Agent Skills Guide
• Agent Middleware Guide
• MCP Guide
• Human In The Loop Guide
• Self-hosted observability
• Offline / air-gapped install
• API Stability and Deprecation Policy
• Upgrading to 1.2.0