memory-informed-design

Memory-Informed Design

Design flow with memory baked in as a non-optional reflex.

Instructions

Step 1: Detect entities

Identify the relevant entity/entities (service, module, subsystem under design) per skills/shared/entity-detection.md.

Step 2: Consult memory (required)

Call pensyve_recall:

• query: a short description of the design question, including secondary entity names where known (e.g., "auth-service jwt signing rs256 hs256")
• entity: primary detected entity
• types: ["semantic", "episodic"] (design benefits most from durable decisions and prior decision-contexts)
• limit: 5

Secondary entities are folded into the query string since the MCP server scopes results by single primary entity only.

Surface one line: Recalled N prior decisions on <entity>. (Skip if N=0.)

Step 3: Recommend with grounding

Shape your recommendation using the recalled decisions. If the user's current question directly contradicts a prior decision, flag it:

Prior decision on <entity> (confidence 0.9): [decision]. Are we revisiting this, or does the current question differ?

Step 4: Capture decision (when it lands)

When the user accepts a design or states a decision ("let's go with X", "we'll use Y"), call the memory reflex:

• Semantic — pensyve_remember with entity: <primary_entity>, fact: "[proactive/in-flight/tier-1] <decision text>", confidence 0.9.
• Episodic — pensyve_observe with episode_id: <session episode_id>, source_entity: "claude-code", about_entity: <primary_entity>, content: "[proactive/in-flight/tier-1] <decision context: alternatives considered, what tipped the balance>", content_type: "text". This enables future "why did we decide X?" queries.

Surface one line: ↳ captured decision on <entity>: <short>.

Step 5: Capture evaluation procedures

If the design process revealed a reusable way to evaluate similar decisions (e.g., "run spike vs. prototype", "check latency first"), capture it as procedural: pensyve_observe with episode_id: <session episode_id>, source_entity: "claude-code", about_entity: <primary_entity>, content: "[procedural] [proactive/in-flight/tier-1] trigger=..., action=..., outcome=...", content_type: "text".

Constraints

• Recall is not optional.
• Observe when a decision lands, not when it's being discussed.
• Budget: one in-flight capture per turn.
• Never store secrets or sensitive architecture details the user has marked private.