Sa Brownfield Extract

SA Brownfield Extract

Use

Use this skill only for SA extract mode during brownfield onboarding Stage 2. The output documents what exists; it does not recommend changes.

Inputs

• docs/archaeology-reports/.md
• Existing codebase/docs named by the report
• Optional scope: dispatch parameter: architecture-only, security-compliance, or a service/module name

Outputs

• Provisional docs/architecture.md
• Observed-decision ADRs only when code irrefutably encodes the decision
• Extracted API/message contract stubs under docs/api-contracts/ when endpoints, RPC operations, or messages are observable. Use service-level OpenAPI for REST, AsyncAPI for brokered messaging, proto for gRPC, GraphQL SDL for GraphQL, and markdown only for legacy/prototype surfaces with an explicit gap note.
• docs/open-issues.md entries with category extract-confirmation-pending
• Worktree plan-update.json

Procedure

1. Read all archaeology reports for the scope. If any required report is INSUFFICIENT, halt and request more archaeology or team-interview input.

• Before extracting, verify the report contains ## Service Boundary & Entry Point Map, ## Route / RPC / Job Trace Matrix, ## Dependency & Call Graph (C3), ## Message Broker / Consumer Logic, and ## API / Message Spec Candidates for the in-scope services.
• If the project has public/internal APIs or brokered messages but those sections are missing or shallow, halt with NEEDS_CONTEXT and request a Stage 1 archaeology re-dispatch with deep route/dependency/contract scope. Do not compensate by writing vague architecture.

1. Map observed services/modules to C4 C1-C3:

• C1/C2 are usually high confidence from repo layout, deploy manifests, and public surfaces.
• C3 is package/module inferred; mark confidence honestly.
• Do not extract C4 Code.

1. Populate architecture sections from observed evidence:

• data models from migrations/ORM/schema files;
• API inventory from Public API Surface + Service Boundary & Entry Point Map + API / Message Spec Candidates;
• per-operation route trace from Route / RPC / Job Trace Matrix;
• dependency/call graph from imports, clients, service registries, Route Trace rows, Dependency & Call Graph edges, and manifests;
• async workflows from Message Broker / Consumer Logic, producers, consumers, topics, state transitions, retry/DLQ configs, and idempotency/dedup evidence;
• cross-cutting concerns from middleware/config;
• NFR posture from metrics when available, otherwise unknown -- measure during pilot.

1. For each discovered API/message, create an extracted contract stub with Status: Extracted, Source: extracted, confidence, observed auth, schemas, and TODO markers for non-introspectable fields.

• REST: create or update docs/api-contracts/.openapi.yaml with openapi: 3.1.0, one operation per observed route, operationId, auth/security scheme, request/response schemas, observed status codes, standard error envelope, idempotency/rate-limit notes, and x-source-evidence file:line values.
• gRPC: create or update docs/api-contracts/.proto with observed service/method/message shapes and TODO markers for unknown field types.
• GraphQL: create or update docs/api-contracts/.graphql with observed query/mutation/subscription names, input/output types, auth notes, and TODO markers for unknown fields.
• Async messaging: create or update docs/api-contracts/.asyncapi.yaml with asyncapi: 2.6.0 or the project's declared version, channels, producer/consumer operation bindings, payload schemas, consumer groups, ack/commit behavior, retry/backoff/DLQ, idempotency/dedup, ordering/partition key, and side effects.
• WebSocket/SSE: document as OpenAPI callbacks/webhooks when the project already uses OpenAPI that way; otherwise create a markdown extracted contract with connection/auth/message/event schemas and lifecycle states.
• Never mark extracted contracts Frozen. FE Dev cannot consume them until BE Dev validates and freezes them during governed implementation or the human confirmation flow explicitly promotes them.

1. Tag every architecture section with Source: extracted | Confidence: high|medium|low|inferred.
2. File one open issue for every Confidence: inferred item.
3. Keep architecture self-contained. Do not put substantive body references back to archaeology reports or source paths.
4. Commit changes, then emit plan-update.json.

Hard Rules

• Document what exists; do not recommend or redesign.
• Do not invent intent.
• Do not omit confidence tags.
• Do not propose new third-party dependencies.
• Do not write SRS or plan content.
• Do not collapse API/message contracts into prose-only architecture. If a route, RPC, or brokered message is observable, write a contract stub under docs/api-contracts/ or halt with a named extraction gap.
• Do not treat a one-sided message observation as complete. A broker flow needs producer, payload, topic/queue, consumer handler, ack/commit, retry/DLQ, idempotency, and side effects when observable; unknown items become TODOs and open issues.
• Commit before signaling done.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: evolplus
• Source: evolplus/talos
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.