Concept image. Normative behavior lives in spec/.
| Document | Purpose |
|---|---|
| Architecture | System boundaries, package map, and execution paths |
| Runtime model | Normative roles, states, routing, usage, and storage boundary |
| Protocol | Normative request, events, errors, usage, and extensions |
| OpenAPI | Machine-readable HTTP surface |
| Roadmap | Completed v0.1 scope and explicitly deferred work |
| Delivery checklist | Current acceptance gates and remaining final-delivery work |
| Transport delivery checkpoint | Transport, lifecycle, SDK and security verification evidence |
| Storage verification | Current persistence, retention, replay, process-crash and container evidence |
| Capability verification | Conservative declarations, configuration, no-call rejection and JSON input mapping |
| Discovery verification | Complete inventory paging, atomic private polling, CLI safety and fresh full local gates |
| Evaluation verification | Bounded scenario checks, profile binding, advisory recommendations and fresh test evidence |
| Publication audit verification | Reachable Git/hosted/media scope, fail-closed scanner fixes and explicit legacy identity warnings |
| Toolchain verification | Native compiler migration, dependency compatibility, clean Linux build and container lifecycle evidence |
| Source-release verification | Audited reproducible archive, tamper rejection and fresh archive installation/startup evidence |
| Document | Purpose |
|---|---|
| Provider adapters | Safe adapter configuration and provider-specific behavior |
| Provider smoke verification | Opt-in live protocol checks, isolated fixtures and sanitized evidence |
| Event storage and recovery | Bounded memory, encrypted journals, durable identity, retention and restart semantics |
| Adding a provider | Public evidence and conformance checklist |
| Gateway integration | Authentication, tenant, quota, and usage boundary |
| Model catalog | Versioned capability snapshots and model discovery diff |
| Model discovery and radar | Account-visible inventories, complete polling, safe baselines and visibility signals |
| Scenario evaluation and recommendations | Bounded assertions, private evidence, exact profiles and single-metric scenario ranking |
| Testing | Reproducible local, SDK, container, and security verification |
| Source release | Fixed-commit source archives, file digests and independent byte-for-byte verification |
| 2026-10-09 hardening verification | Lifecycle fixes, dependency audit, regression results, and CI delivery limits |
| v0.1.0 verification | Release gate results and explicit residual limits |
| v0.1.0 release notes | Shipped scope, verification summary, and platform limitation |
| Document | Purpose |
|---|---|
| Threat model | Shared data-plane assets, threats, and deployment controls |
| Provider evidence | Dated public sources used by clean-room adapters |
| Security policy | Vulnerability reporting and supported security boundary |
| Public contribution policy | Data safety, compatibility, and verification requirements |
| Publication audit | Staged/current/history/hosted disclosure coverage, private policies and failure semantics |
The four editorial images under assets/ were generated through AnyCap and optimized to 1600 x 900
JPEG for GitHub delivery. Each image has a neighboring .prompt.md file containing its public
prompt, generation request ID, processing record, and independent image-read result. Exact protocol
facts remain in Mermaid, OpenAPI, tests, and normative text rather than generated pixels.