System Design Proposal Builder skill

System Design Proposal Builder

Goal: Create a proportionate, evidence-backed target system design that turns requirements into explicit boundaries, contracts, data flow, failure behavior, operations, and tradeoffs. Change only the approved design document; do not implement, audit, or approve the delivery.

Execution contract: The checklist defines completion. Track each item internally as PENDING, PROVEN with evidence, CLEARED with evidence its condition is absent, or UNPROVEN with a gap; reading, delegation, tool failure, a zero exit status, or a self-reported success is not proof; only the observed outcome is. Reconcile after each section. Before returning, resolve all PENDING, count only PROVEN and CLEARED, and apply verdict and approval rules to every gap. Preserve intent, scope, and existing authorization. Continue authorized work; ask only for consequential unresolved choices or required external approval. When no one can answer during the run, state the exact question and apply the skill's verdict for the remaining gap instead of waiting or guessing. Scale depth to material risk without skipping checks. Preserve dependency and safety order; otherwise choose an appropriate verification method. Accept equivalent user or repository evidence; no other skill, named artifact, or complete lifecycle is required. Preserve source requirement and decision IDs. Bind reused evidence to relevant source versions, dirty changes, configuration, and environment; invalidate only affected claims. On continuation, reconcile task, authorization, current state, and unresolved evidence. For long work, return a compact continuation record or update an already authorized artifact; read-only skills do not persist it. Distinguish artifact readiness, verified behavior, and external-action authority. Prepare authorized work before required approval. If blocked by an instruction, cite its exact source and unresolved boundary; do not invent approval gates from caution.

Tool Routing

Need	Preferred capability	Fallback
Requirements and constraints	Approved requirements, baseline, decisions, and direct stakeholder input	Mark material gaps and ask the smallest decision question
Current implementation and conventions	Repository search, manifests, entrypoints, and architecture artifacts	Treat as greenfield only when the user or repository establishes that fact; otherwise mark current state UNKNOWN and return INCOMPLETE or BLOCKED when the gap can change boundaries, compatibility, or migration
External capabilities and limits	Current official documentation and specifications	Mark claims UNVERIFIED; avoid vendor-dependent commitment
Estimates	Reproducible arithmetic from sourced workload assumptions	Use ranges and sensitivity; never present estimates as measurements
Document mutation	Minimal patch to the approved target-design artifact	Return BLOCKED if scope or path is unsafe

Use patterns as candidate solutions, not goals. Introduce infrastructure only when a requirement, failure mode, ownership boundary, or measured horizon pays for its lifecycle cost.

Artifact Rules

• Reuse a clear target-design document; otherwise use docs/architecture/target-design.md.
• Read available baseline, current-state, decision, interface, diagram, and migration artifacts by path; none is mandatory.
• Label facts, assumptions, estimates, proposed decisions, and unresolved choices separately.
• Compare credible alternatives for consequential decisions, including the simplest feasible option; when constraints permit only one, document why rather than inventing another.
• Prefer reversible choices and the simplest topology fitting the system. For a new application, consider a modular monolith before independent services; do not force that shape onto libraries, plugins, or an established topology.
• Do not silently change an accepted decision; record the conflict and required governance action.

Checklist

1. Frame the Design

• Resolve business outcome, actors, journeys, scope, non-goals, horizon, readers, language, and the approved canonical destination before editing.
• Read repository instructions and inspect relevant architecture artifacts and current implementation.
• Extract functional requirements and measurable quality drivers, preserving their source and status.
• Identify architecture-critical unknowns and ask only questions whose answers change the target shape.
• Return BLOCKED when a required business boundary or safety constraint cannot be responsibly assumed.

2. Estimate Before Choosing Components

• Estimate average and peak request or event rates, concurrency, payload and bandwidth, storage growth, retention, and recovery volume where relevant.
• Show formulas, ranges, growth horizon, and assumptions; identify the variables that can reverse a choice.
• Identify likely first bottlenecks and explicit thresholds for deferred scaling mechanisms.
• Separate availability, latency, durability, consistency, security, cost, and operability requirements from implementation preferences.
• Reject speculative scale and list complex mechanisms intentionally deferred.

3. Define Domains, Data, and Contracts

• Map business capabilities, domains or modules, ownership, invariants, and allowed dependency direction.
• Define systems of record, data models at architecture depth, lifecycle, retention, consistency, and transaction boundaries.
• Define public APIs, events, commands, schemas, errors, idempotency, ordering, versioning, and compatibility expectations.
• Define trust boundaries, identities, authorization, sensitive data, secrets, abuse controls, and audit needs proportionate to risk.
• Keep framework and vendor details outside the core model unless they are genuine constraints.

4. Build HLD and Critical LLD

• Describe system context, deployable units, stores, queues, external systems, responsibilities, and labeled data flows.
• Trace success, overload, dependency failure, partial failure, retry, timeout, degradation, recovery, and cancellation for critical journeys.
• Deep-dive only components whose correctness, scale, security, or reversibility risk warrants implementation-level detail.
• Define observability, SLI measurement points, health, deployment strategy, rollback, backup, and operator actions.
• Define ownership, team impact, cost drivers, and operational burden for the proposed topology.

5. Decide and Validate

• Compare credible alternatives against requirements, estimates, failure behavior, complexity, cost, migration, and future triggers.
• State selected and rejected options with consequences, sensitivity points, and assumptions that would reopen the decision.
• Identify significant decisions that deserve their own compact decision records without requiring another workflow.
• Define architecture acceptance evidence appropriate to each material driver: contract checks, load/failure experiments, security validation, recovery proof, or observability signals. Specify prerequisites and pass criteria; do not execute them during design.
• Outline current-to-target implications and compatibility needs without expanding into a full implementation plan.

6. Write and Report

• Write context, drivers, estimates, domains, contracts, HLD, critical LLD, failure and operations model, security, alternatives, decisions, validation, open questions, and evolution triggers.
• Preserve existing content outside the approved scope and link shared artifacts only by document path or title.
• Re-read the proposal for unsupported facts, hidden decisions, mixed abstraction, and unjustified machinery.
• Demonstrate that each material design boundary is implementable: link source requirements to contracts, data, failure behavior, unresolved choices, and acceptance evidence; distinguish a ready proposal from an accepted decision.
• Use READY only when the design is decision-complete enough for implementation planning; use INCOMPLETE for material but solvable gaps; use BLOCKED when required intent, evidence, authority, or destination is unavailable.

Self-Check

• Reconcile before returning. Check item-level evidence, requirement coverage, contradictions, scope, verdict, and applicable cleanup. Correct the report or authorized artifacts. Reuse valid evidence; do not automatically rescan the repository or rerun successful commands. Repeat checks only for relevant changes, failures, or unresolved evidence. Disclose remaining gaps.

Output Contract

Report in the user's language, in this order; label all five fields and state each fact once. Use controlled plain language: one fact per sentence, usually under 20 words, active voice, and one term per concept, with no synonyms for verdicts, IDs, or states. Small results may use one line per field; omit empty tables and do not copy linked artifacts:

1. Result: The exact skill-specific verdict token first, then the supported outcome.
2. Scope: Reviewed/changed scope, exclusions, baseline, and material assumptions.
3. Evidence: Skill-specific fields below; distinguish facts, inferences, and unverified claims. Link artifacts; use tables when useful.
4. Verification: Checks/results, unavailable evidence, and applicable cleanup/external state.
5. Completion: Checklist: X/Y complete; Incomplete: None or each UNPROVEN item's reason, outcome impact, and exact next action; residual risks and required decisions.

Skill-specific evidence: Artifact path; requirements, estimates, boundaries, contracts, HLD/critical LLD, selected decisions, alternatives, evidence, and reopen triggers. Summarize validation/transition needs, compatibility, rollout, rollback, observability, and only unresolved choices that affect implementation planning. When the design has more than a few components and the host renders Markdown diagrams, add one Mermaid boundary diagram; keep the text complete without it.