Focus: Design the public API surface — the contract that consumers will depend on. This is load-bearing work because once published, changing the public surface breaks every consumer. Decisions here set the semver policy and dictate how painful every future release will be.

Produces: API Surface document with:

• Public entry points — every function, class, type, or module exported from the library, with signatures
• Error model — how errors are returned (exceptions, result types, sentinel values), and what error classes exist
• Extension points — how consumers customize or extend behavior (hooks, subclassing, middleware)
• Semver policy — what constitutes a breaking change in this library's judgement (signature changes are obvious; behavior changes less so)
• Stability tiers — if the library has experimental/stable/deprecated APIs, define the tiers and their guarantees
• Naming & conventions — module layout, export style, idiomatic usage patterns consumers will see first

Reads: Discovery document, ecosystem conventions for the target language, existing code in the repo.

Anti-patterns (RFC 2119):

• The agent MUST NOT design internal implementation details — only what consumers will see
• The agent MUST NOT expose framework primitives that leak into consumer code (e.g., returning internal classes)
• The agent MUST prefer small, composable public APIs over large, monolithic ones
• The agent MUST specify what consumers can rely on and what they cannot (internal namespace conventions, underscored names, etc.)
• The agent MUST NOT design for hypothetical future consumers — design for the users identified in discovery

Focus: Decompose the intent into verifiable units. Each unit should be scoped to a single bolt and have concrete, testable completion criteria.

Produces: Unit specifications with:

• Clear scope — what this unit delivers and what it explicitly doesn't
• Dependencies — which other units must complete first (feeds the DAG)
• Completion criteria — verifiable conditions, each tied to a command or test
• Unit type — research for inception units

Reads: Discovery document, API Surface document.

Anti-patterns (RFC 2119):

• The agent MUST NOT create units with vague criteria like "implementation complete"
• The agent MUST ensure the unit DAG is acyclic
• The agent MUST scope each unit to a single bolt
• The agent MUST NOT skip API surface units — consumers depend on that contract

Focus: Understand the problem this library solves, who consumes it, and what the competitive landscape looks like in this ecosystem. Libraries live or die by adoption — establish who will use it, why they'd pick it over alternatives, and what consumer experience the library needs to deliver.

Produces: Discovery document with:

• Problem statement — what consumers can't do today, and what this library unlocks
• Target consumers — who integrates this (backend devs, CLI users, other libraries, etc.) and at what technical sophistication
• Ecosystem survey — existing libraries in this space, what they do well, where they fall short, and licensing/governance differences
• Adoption signals — how consumers will find and evaluate this library (package name conventions, docs expectations, example style)
• Non-goals — what this library explicitly will NOT do, to bound scope

Reads: Intent problem statement, existing consumer code in the repo, competitor library docs and READMEs.

Anti-patterns (RFC 2119):

• The agent MUST NOT propose the API surface here — that's the api-architect's job
• The agent MUST NOT skip ecosystem survey — libraries fail most often by ignoring what consumers already use
• The agent MUST ground the discovery in real consumer needs, not hypothetical users
• The agent MUST identify non-goals explicitly — scope creep kills libraries