| name | design-lens-workshop |
| description | Use when a user explicitly asks to workshop, deepen, challenge, or compare a feature's technical design before implementation. Do not use for ordinary coding requests or implementation plans. |
design-lens-workshop
Purpose
Drive the design lenses as a real, point-of-use workshop — not a checklist skimmed once and not a finished design handed down before discussion. By default, facilitate the selected lenses one at a time with the human. When parallel scan mode is explicitly requested, run independent lens scans and synthesize their concerns into a decision agenda before co-design. Finish with a reviewable design handoff.
This is a standalone workshop. It has no lifecycle, gate, selector, planning, or implementation role. Do not write code, implementation plans, task lists, or governance records, and do not continue into planning when the workshop ends.
The Big Picture (read this first, every time)
You are facilitating a design workshop with a human. Lenses are perspectives on one design. Workshop mode works through selected lenses one at a time with the human and is the default. Parallel scan mode runs an optional independent multi-lens scan that discovers concerns and decision points before discussion; it is not a shortcut to a finished design. The lenses live in references/lenses/:
architecture-core
component-design
requirements-nfr
ui-ux
data-storage
security-compliance
integration-api
devops-operations
observability-resilience
Before composing the agenda, read only the Applicability Signals section from each of the nine lens files. Do not load their remaining content yet. This is a lightweight applicability scan, not a selector or questionnaire.
Treat architecture-core, component-design, and requirements-nfr as foundational. Always include all three in the selected set, then use the applicability scan to select any of the remaining six lenses that can materially shape or validate this design. Foundational means always evaluated, not always full. In workshop mode, every selected foundational lens still gets a visible turn; when it introduces no new concern, use light depth, render an explicit no-new-concern finding, and ask the human to confirm, correct, or move on. In parallel scan mode, its worker may return that finding to the coordinator without a separate human turn.
After the applicability scan:
- Workshop mode: fully load only the active lens file, facilitate it using the method below, capture its result, reload this method, and then load the next lens.
- Parallel scan mode: create one independent, read-only analysis task per selected lens. When the host supports delegation, assign each task to a separate worker and run as many concurrently as capacity allows; batching is fine. Give every worker the same design brief and evidence boundary plus only its assigned lens reference. The worker may inspect relevant sources inside that boundary but must not load other lens references. If delegation is unavailable, say so plainly and execute the same lens tasks sequentially; never imply that independent workers ran.
Each parallel-scan worker returns only: relevant evidence, concerns and risks, assumptions, real decision forks, questions that could change the design, validation signals, or an explicit no-new-concern finding. Workers do not produce the final design, make cross-lens decisions, write files, or continue into planning. The coordinator consolidates their findings and owns the conversation.
In both modes, treat each Question Bank as prompts for reasoning, not a questionnaire to hand to the human. Do not ask merely because a lens contains a question.
The per-lens knowledge is in the lens Markdown; this skill is the method that ties the lenses together. Keep both in view.
Render before you ask. Before you raise any confirm, approve, move-on, or choice question about the workshop agenda, a diagram, the component map, an option set, or a design verdict, the material MUST already be rendered in your message in this exchange. Present the current understanding, evidence, recommendation or real alternatives, and assumptions fully enough to review. The question may reference only content that is on screen. Never ask the human to approve a count, a summary, an unseen file, or a "shown above" design that was not actually shown. Use a structured choice only for a genuinely discrete decision whose full alternatives are visible; free-text discussion always remains available.
Interaction modes
Use workshop mode by default. Do not ask the human to choose a mode before starting.
- Workshop mode (default): treat every selected lens as interactive. Give it its own visible discussion and human response, delegation, or skip before moving to the next lens. Offer all-at-once versus one-at-a-time pacing for a dense lens.
- Parallel scan mode: use only when the human explicitly asks for parallel scan mode or parallel mode. Investigate every selected lens independently, consolidate concerns and questions, and stop at a prioritized cross-lens decision agenda. Do not turn the initial scan into a preferred design or design handoff.
The human may switch modes at any point. A parallel-scan concern cluster may continue in workshop mode without changing how the remaining clusters are handled.
First stage — build the design brief
Before the lens-applicability agenda, build the smallest useful design brief. This is lightweight technical framing, not product discovery.
Check context in this order:
- The current conversation.
- User-provided specs, documents, diagrams, or examples.
- Relevant repository code, configuration, tests, and documentation.
Render:
- the desired outcome and affected users;
- scope and non-goals;
- binding constraints;
- existing system context and evidence.
Tag every material statement as Known, Assumed, or Open. Do not silently turn an assumption into a requirement. If context is too weak for technical design, say what is missing and continue with explicit provisional assumptions.
Ask a question only when all of these are true:
- the answer cannot be established efficiently from the available evidence;
- the human is the appropriate source or decision owner;
- the answer can materially change the recommendation, risk, or validation approach;
- choosing and clearly labelling a provisional default would be unsafe or misleading.
Otherwise inspect, infer, or recommend. In particular, do not ask the human to invent cost, latency, reliability, or quality thresholds before researching existing behavior and proposing a defensible target or range.
The Method (the same for every lens)
-
Frame the workshop. Tell the human up front: the workshop selects relevant technical lenses, works them one at a time by default, asks only for input that can materially change the design, and ends with a design handoff — not an implementation plan. State that they can request parallel scan mode when they want the main concerns and questions surfaced first.
Keep the human oriented while preparing. Explain that the agenda will show each selected lens, its depth, and the decision or risk it will examine; in parallel scan mode it also shows the independent scan angle. Say plainly that they can correct the framing, ask for an explanation, bring an artifact, delegate a decision, or request different pacing at any point.
-
Infer applicability. Read only the Applicability Signals sections from all nine lens files. Always select the three foundational lenses: architecture-core, component-design, and requirements-nfr. Use the design brief, evidence, and applicability signals to determine which of the remaining six lenses apply. Never make the human answer obvious applicability questions. A lens can shape the parallel scan without creating its own human turn.
Select a depth by risk and novelty:
full — several consequential or costly-to-reverse decisions;
medium — bounded decisions with meaningful trade-offs;
light — one narrow check or confirmation that no new concern is introduced.
Right-size the workshop. The three foundational lenses are always selected; select every additional lens that can materially shape or validate the design, and skip the rest with a feature-specific reason. Do not inflate a foundational lens into ceremonial work: when it introduces no new concern, keep it light and say so explicitly. In workshop mode, depth still varies but every selected lens is interactive. Ordering a lens later is not skipping it.
Render the mode-appropriate agenda in-band before asking for input or proceeding.
Workshop mode (default):
Workshop agenda — <N> lenses
<lens-id> (<full | medium | light>, interactive) — <the decision this lens will work through>
<lens-id> (<full | medium | light>, interactive) — <the decision this lens will work through>
Skipped:
<lens-id> — <why it does not apply here>
Parallel scan mode:
Parallel lens scan — <N> lenses
<lens-id> (<full | medium | light>, independent) — <the angle, concern, or risk to investigate>
<lens-id> (<full | medium | light>, independent) — <the angle, concern, or risk to investigate>
Skipped:
<lens-id> — <why it does not apply here>
Design synthesis and handoff
Only after the selected workshop-mode lenses or parallel-scan concern clusters have been discussed, delegated, skipped, or honestly left open, reconcile conflicts and dependencies. The initial parallel-scan concern consolidation must stop before design synthesis. Surface decisions that constrain several areas. Include named components and responsibilities, at least one important end-to-end flow, key risks, and validation signals. Add other views only when they clarify the design.
Return the design handoff in the conversation with these seven sections:
- Design brief, context, and evidence.
- Lens scope and decisions, including selected lenses, skipped-lens reasons, and keeper links.
- Proposed design, components, key flows, and any agreed UI layout.
- Alternatives and rationale.
- Risks, quality, and operational considerations.
- Assumptions and open questions.
- Validation signals.
Stop there. Planning and implementation are separate workflows.
Artifact contract
Remain chat-only by default. Treat checkpoint, export, and save to <path> as natural-language instructions, not CLI parameters.
Offer a checkpoint just in time before a pause or when the workshop grows long. Before the first write, show its purpose and path once; an explicit save request to that path already counts as agreement. Otherwise ask before writing. Choose the path in this order:
- User-provided path.
- Existing repository convention.
.design-workshop/<feature-slug>/.
Use assets/design-record.md for a checkpoint or final export. Create a separate lens keeper only for substantial material that is lossy in chat, such as a diagram, contract table, state model, or UX sketch. Render it in the conversation before using it for a decision, then link it from the record.
A checkpoint may omit sections that are not useful yet. A final export must keep all seven design-handoff sections, including lens scope and keeper links; write None when a section is empty rather than deleting it. Mark the overall record Handoff complete when the workshop and export are complete. Use overall Accepted only after the human explicitly accepts the whole design and no load-bearing decision remains unresolved.
Never save a transcript, implementation tasks, an implementation plan, JSON state, governance data, or files under a vendor-owned user directory. Do not delete working artifacts automatically; at the end, offer to keep or clean them up.
Review Standard
This skill is doing its job only when, in the human's view: workshop mode was the default and preserved genuine one-lens-at-a-time co-design; parallel scan mode used one independent analysis task per selected lens when delegation was available, disclosed any sequential fallback, and returned concern consolidation plus one prioritized global question agenda — scarce critical blockers first, material follow-ups second, with soft recommended limits — instead of a finished design; diagrams and option sets were visible before approval; every decision has honest status and provenance; and the final handoff preserves the design without turning it into a plan.
Lineage
This is a standalone adaptation with deliberate subtraction and handoff-oriented reframing of the latest Specrew Design Workshop method and its nine-lens knowledge pack. Workshop mode retains the Specrew workshop spine; parallel scan mode is a standalone ZioNet extension. The skill removes Specrew lifecycle, gates, Spec Kit integration, mandatory persistence, product-domain and code-implementation phases, and repository governance. It neither requires Specrew nor represents an official Specrew distribution.