Design Blueprint

Required Skills

Read and apply in order before Step 1:

1. framework:knowledge-priming -- Load the project knowledge base so every decision grounds in the real project. (always)
2. framework:context-anchoring -- Create or load the feature's living context doc (Create / Load / Enrich behaviors). (always)
3. framework:learning-harvest -- Load prior operational learnings at session start; harvest new ones at session end. (always)
4. framework:collaborative-judgment -- Surface genuine judgment calls as structured options instead of silently assuming. (always)
5. framework:design-first -- Owns the 5-level methodology. Its Entry Assessment, Complexity Calibration, Simplicity Check, and Level Completion Protocol govern Step 2. (Step 2)
6. framework:architecture -- Validate components, layers, dependency direction, and boundary rules (design mode). (Levels 2-4)
7. framework:domain-driven-design -- Model aggregates, entities, value objects, events, and contracts (design mode). (Levels 2-4)

Additional Instructions

STOP: Before Step 1, read additional_instructions.design_blueprint (one line or a list) from .lattice/config.yaml. No file or no key → skip silently.

When lines are found:

1. Show them verbatim under "Additional instructions (from config):".
2. Highest priority — apply them for the whole session, including on resume. On any conflict with anything else in this skill — steps, gates, STOP rules — or with its Required Skills, standards documents, or operational learnings, the config line wins.

Workflow

Step 1: Establish Context

1. Run framework:learning-harvest Load behavior. Focus hint: "design session — focus: design patterns, reliability, structural health".

2. Set up the feature's living doc with framework:context-anchoring:

   • Discover: run context-anchoring Document Discovery — scan the context base directory (per the atom's Config Resolution) for an existing anchor doc matching the feature name or frontmatter.
   • Found → Load behavior. Present the structured acknowledgment: feature name, status, decision count, open questions, constraints. Then run the resume check below.
   • Not found → Create behavior. Confirm the feature name, summary, and requirement doc link with the user before creating. Then begin Step 2 — the Entry Assessment sets the entry level.

3. Resume check (when a doc was found) — derive the earliest incomplete step from the doc itself. STOP: Never re-walk agreed work:

   • status: approved → design is finished. Say so and stop; suggest /code-forge.
   • No sections starting ## Design: Level → begin Step 2 — the Entry Assessment (and its spec rule) sets the entry level.
   • Some levels persisted → summarize the approved levels briefly, then resume at the first missing level at or after the recorded entry level (the [Entry] Decisions Log entry; older docs without one → treat entry as Level 1).
   • Every level from entry through Level 4 persisted, but no ## Design Summary, or status ≠ approved → go directly to Step 3.

4. Requirement constraints: read requirement_doc from the context doc frontmatter.

   • Absent → skip.
   • Local path, unreadable → STOP: "Requirement doc not found at [path]. Verify before continuing."
   • Local path, readable → read it and extract ## Technical Constraints. Treat as non-negotiable — same authority as architecture rules. Surface to the user before the first level is presented.
   • External reference (URL, ticket ID, or other non-local-path identifier) → resolve via a connected MCP tool if one can. If none is connected or the fetch returns nothing, ask the user to paste the current constraints — expected, not an error.
   • Either path resolved to the spec itself (not pasted constraints only) → Step 2's spec rule applies.
   • Conflict during design → surface via framework:collaborative-judgment. The user decides; record the change back in the requirement doc's ## Technical Constraints if local, or in the Decisions Log if external — this molecule never writes to an external system.

5. Write the back-link: if requirement_doc resolved to a readable local file, add to its ## Links section (create it at the end if missing): - Design: [{context-doc-filename}]({path from the requirement doc to this context doc}). Several context docs may link one requirement (e.g. one per platform) — one line each. One discrete file edit; skip if this context doc's link is already present.

Step 2: Walk the Design Levels

Run design-first's Entry Assessment first: state the proposed entry level from its Complexity Calibration table and wait for confirmation. Record the confirmed entry level in the Decisions Log: [Entry] Start at Level N (name) — rationale. If key use cases or success criteria are unclear, surface them via framework:collaborative-judgment before producing the first level output.

Spec as Level 1: when the context doc has no ## Design: Level section yet and requirement_doc resolved to the spec itself, the spec is Level 1. Do not re-validate it — spec quality belongs to requirement-forge.

• Propose entry at Level 2 or the calibrated level, whichever is later, and offer: "Level 1 comes from the linked spec. Starting at Level [N]. Want a capabilities walk-through first?" STOP — do NOT advance until the user answers.
• User wants the walk-through → enter at Level 1, using the spec as input.
• Otherwise, save ## Design: Level 1 -- Capabilities as one line: Taken from [{spec-file}]({path from this context doc to the spec}). Not re-walked. — compute the relative path from the actual locations of both files; an external spec gets its reference instead. Do not copy the spec.

Drive the levels sequentially from the confirmed entry level through Level 4 via framework:design-first. Complexity Calibration sets how deep each level goes; it never removes a gate or skips persistence.

Gate (every level) — follow design-first's Level Completion Protocol: present the level output with its targeted gating question, then STOP — do NOT advance until the user explicitly confirms, not on silence, not on ambiguity.

Persist (after every approval, before advancing) — use framework:context-anchoring Enrich to write into the context doc:

1. The approved output as a clean structured summary under ## Design: Level N -- {Name}, same format as presented (numbered list L1; component table + diagram L2; sequence/flow L3; typed interfaces L4). Persist diagrams as Mermaid.
2. One Decisions Log entry per decision: [Level N] Chose X because Y. Rejected: Z.
3. Constraints identified during the discussion (non-negotiable boundaries that emerged).
4. Open questions surfaced but unresolved.

STOP: Do not present the next level until these writes are done.

Judgment calls: when applying architectural atoms at any level, surface genuine design judgment calls immediately via framework:collaborative-judgment — never batch them to the end of a level.

Evidence rule (Level 2): before presenting components, quickly explore the codebase and map each proposed component to the existing modules/packages it extends, wraps, or modifies — or mark it new. Present the mapping with the components. Never invent a parallel structure that ignores what exists.

Level-specific applications:

• Level 1 (Capabilities): numbered user-facing capabilities, max 5, no technical detail (per design-first).
• Level 2 (Components): challenge each component before approving — does it need to exist? One known implementation, one caller, or an unconfirmed problem → inline it or defer. Then validate in design mode: framework:architecture (layer mapping, dependency direction, boundary clarity) and framework:domain-driven-design (aggregates, entities, value objects; domain vs infrastructure placement).
• Level 3 (Interactions): framework:architecture — data flows follow the loaded patterns; boundary-crossing rules respected. framework:domain-driven-design — cross-aggregate communication uses domain events / eventual consistency.
• Level 4 (Contracts): framework:domain-driven-design — repository interfaces, value object types, aggregate root boundaries reflecting the tactical choices from earlier levels. framework:architecture — boundary-data rules and interface ownership respected. Every Level 3 interaction maps to at least one interface.

Regression rule: if the user reopens an approved level, re-run that level's gate. On re-approval, mark every downstream persisted level section stale ("stale — pending re-approval after Level N change") and re-present them for confirmation before Step 3. STOP: Never leave contradictory approved sections in the doc.

Early exit: if the user wants to stop or shortcut the design, follow design-first's Mid-level exit. Persist whatever was approved and leave status as draft — a partial doc is a valid outcome.

Step 3: Finalize Blueprint

After Level 4 is approved and persisted:

1. Verify completeness and consistency: the context doc must contain all four level sections plus every decision made during the design. Enrich anything missing now. Then check:

   • Every Level 3 interaction maps to at least one Level 4 interface.
   • Every Level 4 interface is owned by exactly one Level 2 component (a shared type is owned by its defining component). Fix any gap through the affected level's gate — never silently.

2. Check requirement spec drift: read requirement_doc from the context doc frontmatter.

   • Absent → note in Design Summary: "No requirement doc — drift check skipped."
   • Local path, unreadable → STOP: "Requirement doc not found at [path]. Verify before continuing." (A broken local path is an error.)
   • External reference, unresolvable (no connected MCP tool, or the fetch returns nothing) → do not STOP — expected, not broken. Ask the user to paste current constraints/scenarios if a comparison is wanted, or note in Design Summary: "Requirement doc is external and unavailable this session — drift check skipped."
   • Resolved (local file read, external fetch succeeded, or user pasted constraints) → compare L4 contracts against Scenarios/ACs and ## Technical Constraints. Present each divergence as [field/behavior] — changed from [X] to [Y]. Reason: [from Decisions Log], or "L4 consistent with requirement spec — no overrides" if none. Ask: "Record this in the requirement doc?"
   • STOP: do not write to requirement_doc until confirmed. Confirmed and local → write each finding into the requirement doc's ## Links section as - Design override: [field/behavior] — changed from [X] to [Y]. Reason: [...], or - Design alignment: L4 consistent with requirement spec — no overrides. if none. Confirmed and external → this molecule never writes to an external system; record the findings in the Design Summary instead. Declined → note in Design Summary: "Drift check results not written to requirement doc — see Decisions Log."

3. Write the design summary: use framework:context-anchoring Enrich to add a ## Design Summary section containing components and layer assignments, key contracts and interfaces, architectural constraints, domain model decisions (if applicable), and open questions resolved during design.

4. Set approved status: write status: approved into the context doc frontmatter. STOP: discrete file edit — not prose. Without it, code-forge will not proceed. STOP: never write status to requirement_doc — the requirement's status belongs to whoever manages it (a human, or an external system); this molecule manages only its own context doc.

5. Log the completion decision: "Design approved at Level 4. Status set to approved — ready for implementation." Present the summary to the user as confirmation.

6. Harvest learnings: run framework:learning-harvest Harvest behavior. Session context: "design session — architectural decomposition and contract definition". Synthesize and propose cross-cutting patterns from this session — decomposition approaches, architectural trade-offs, scope decisions that could inform future designs. The user confirms what enters the document. STOP: run this before the next bullet — do not jump straight to the /code-forge suggestion.

7. Design complete. Do NOT proceed to Level 5 (Implementation). Suggest the user invoke /code-forge when ready to begin coding against the approved design.