Foundation specifications.

Defines a portable, vendor-neutral manifest format that describes how a coding agent stores, retrieves, and exchanges memory across runtime boundaries. The goal is for an agent built on one platform to hand off its working context to an agent on a different platform without loss of meaning.

Specifies the minimum set of core fields that every software factory MUST publish to describe itself: identity, capabilities, supported agent types, telemetry endpoints, and conformance claims. The Factory Manifest is the document an external observer reads to understand what a factory is and what it does.

Defines a structured issue representation that both humans and AI agents can read, write, and decompose. ARIF supersedes the informal Markdown body of typical issues with explicit fields for dependencies, success criteria, agent ownership, and decomposition state.

Defines a portable metadata schema for design system components — naming, props, variants, slots, accessibility intent, and intended usage — so that AI agents and design tools can reason about a design system without parsing its rendering layer.

Extends the conventional human-readable Architecture Decision Record with strict YAML front-matter and a small grammar for stating the decision, its forces, and its consequences in a form agents can query, link, and check for conflicts.

Defines a standard envelope for test results emitted by factory pipelines: identity, status, evidence, environment, lineage, and severity. UTRE is transport-neutral and accumulates across stages of a single change.

Defines how a coding agent declares its capabilities, tool requirements, supported protocols, and trust boundaries to an orchestrator. Capability negotiation is mandatory; static configuration is OPTIONAL.

Specifies the output schema for an automated or human-augmented code review: findings, severity, anchor (file/line/symbol), category, suggested fix, and disposition. Multiple SROS outputs from different reviewers can be merged into a single review.

Defines the manifest a software factory emits to a deployment system to release a build: identity, environment targets, approval signatures, rollback criteria, and verification probes. Manifests are signed and replayable.

Defines a starter set of benchmarks for evaluating software factories on completeness, correctness, speed, cost, and safety. Benchmarks are reproducible, public, and version-pinned so claims of "our factory is better at X" can be compared apples-to-apples.

Defines how an agent declares which pieces of its context expire when, why, and what to do when an expiration is reached (drop, summarize, archive, escalate). Without an expiration policy, agent memory grows unbounded and becomes unsafe.

Defines a mapping between Factory Manifest events (SFF-RFC-0002) and the [OpenLineage](https://openlineage.io) specification so factories can publish to existing OpenLineage consumers without parallel instrumentation.

Defines how one agent invokes a tool that is owned by another agent, including capability handshake, authorization, argument schema, return envelope, and audit propagation. Composes with the Agent Capability Manifest (SFF-RFC-0007).