- Home
- Skills
- APIs & Backend
- Software Architect Specialist
Works with the AI Agents you already use
Software Architect Specialist
Design beyond the happy path. Turn software goals into clear boundaries, data and API contracts, failure behavior and a buildable implementation plan.
$5
Software Architect Specialist
Example session with this skill installed
Design a cancellation-safe export feature for our small invoicing app. Current system, supplied description only: one web application and relational database; POST /exports starts a job; a worker writes a CSV to object storage. A user can accidentally submit twice after a timeout. An authenticated user must see only their own exports. Retain completed files for seven days. Cancellation should stop work where possible; if completion races with cancellation, the download must remain unavailable after cancellation is accepted. We have no repository access, traffic measurements or existing worker contract here. Produce an implementable architecture specification, not code. Reuse the current application unless there is a concrete reason to add a service.
- Read your context and instructions
- Compiled the software architect specialist
Illustrative preview: this representative editorial result is not a recorded runtime session.
ArchitectureSpec — cancellation-safe invoice exports
task_id: software-architect-example-1; artifact_id: ArchitectureSpec-EX-1; version: 0.1.0. IDs identify this representative artifact, not a runtime execution.
Status: DRAFT. Producer: Software Architect; recipient: implementation owner. Objective: one export per intentional request, owner-scoped access and cancellation that remains authoritative. Grounding: the supplied system description; base_ref unknown. No code or deployment was inspected.
Recommendation
Keep the existing application, database and worker. Add an export lifecycle and an atomic request-identity contract. A separately deployed export service would add network and operational failure without a stated isolation or scaling need. Revisit that choice only when measured resource interference or independent ownership justifies it.
Rules and boundaries
REQ-1: request identity is scoped to authenticated owner plus operation. The client supplies a fresh identity for each intentional export and reuses it after an uncertain timeout. The server stores identity, canonical parameters and job ID atomically with job creation. A repeat with matching parameters returns that job; a parameter mismatch is rejected. Identity retention must cover the seven-day job lifecycle. This is a proposed API contract, not an inspected existing one.
REQ-2: resolve owner from trusted authentication, never from client-submitted owner_id. Every status, cancellation and download lookup checks the authenticated actor against the stored owner. Cross-owner lookups disclose neither metadata nor storage URLs. Storage remains private.
REQ-3: proposed states are queued, running, completed, cancelled, failed and expired. Cancellation sets a durable cancellation marker in a transaction. Completion may attach an object reference only if cancellation has not been accepted; conditional state change resolves the race. If upload already occurred, cleanup removes that object, but download authorization denies it immediately. Cancellation acknowledgement therefore depends on the durable marker, not on worker interruption.
REQ-4: status returns state, safe progress if supported and permitted next actions. Failed work is retryable only through a defined action; blind worker retry must not create a second export. The worker checks cancellation before expensive work and before publishing completion. Unknown upload outcomes are reconciled by recorded job/object identity.
REQ-5: expiry denies download at the server after seven days and schedules object deletion. Decide whether job metadata survives file expiry before implementing its retention policy; that policy is not supplied.
Acceptance and failure signals
AC-1: concurrent submissions with the same owner, identity and parameters create one job; any second job fails the contract. AC-2: distinct intentional identities may create distinct jobs even with identical parameters. AC-3: a matching retry after a simulated response timeout returns the original job. AC-4: another user cannot read, cancel or download the job. AC-5: both completion/cancellation orderings leave download denied after accepted cancellation; an accessible object fails the invariant. AC-6: expiry denies download and cleanup can be retried without reviving the export. All checks are proposed, not executed.
Delivery and recovery
Slice 1: inspect current callers/worker, add lifecycle and identity storage, then prove AC-1–4. Slice 2: add cancellation/completion race enforcement and cleanup, prove AC-5. Slice 3: expiry enforcement and reconciliation, prove AC-6. If existing clients omit identities, a coexistence policy is required before changing POST semantics. Back out new creation through a feature control while retaining lifecycle records for outstanding jobs; a code rollback does not undo existing object or data effects.
Next owner/action: implementation owner inspects current transaction and worker boundaries and closes metadata retention and legacy-client coexistence. Independent contract design can proceed. Those gaps keep the affected delivery slice DRAFT; this specification is not implementation or release approval.
Connects securely to your tools. The creator never sees your data.
About this skill
A stack diagram does not tell an engineer what to build when the request times out.
Software Architect Specialist equips your AI to turn a software goal into an implementable design. It traces the current system, compares consequential options and closes the contracts between components before coding begins. Get the design decisions, failure behavior and acceptance checks that make a technical handoff useful.
Use it for a new feature, integration, refactor or defect that crosses technical boundaries. Bring the requested outcome, current repository or system description, business rules and constraints. The method scales from a compact local change to a multi-module design without adding infrastructure for hypothetical future needs.
Your architecture packet can include
- System boundaries and ownership of data, rules and operations.
- API and data contracts with validation, errors and lifecycle behavior.
- Decisions with alternatives, consequences and revisit conditions.
- Testable delivery slices, migration steps and recovery plans where needed.
For an export feature, it can work through duplicate submissions, owner-only access, retention and the race between cancellation and completion. An engineer receives the intended behavior across those boundaries, including the decisions still needed before implementation.
The complete standalone method, design references and specification format are included. No other squad skill is required. Supply your own AI access and system materials. It produces architecture and a handoff; repository access improves grounding, while actual implementation and runtime verification remain separate work.
How to use
Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scanned
Verified clean 16 days ago
- Passed all security checks, Safe to install