avifenesh/decision-support

Decision Support

Triggered when the user's question involves a choice, tradeoff, or recommendation.

Before answering, gather context from memory and past sessions so the recommendation is grounded in what Avi already knows and has decided before.

Step 1: Extract the Decision Question

Identify the core choice being made:

• What are the alternatives?
• What domain is this in (technical, personal, strategic)?
• What constraints or preferences might apply?

Step 2: Retrieve Prior Context (run all three in parallel)

Past decisions (cairn.deepSearch):

cairn.deepSearch("<decision topic> decision choice tradeoff")

Look for: previous decisions on this exact topic, related decisions, documented preferences.

Journal history (cairn.journalSearch):

cairn.journalSearch("<topic keywords>")

Look for: sessions where this topic came up, outcomes of previous choices, context that led to past decisions.

Facts and preferences (cairn.searchMemory):

cairn.searchMemory("<topic> preference constraint requirement")

Look for: hard rules that eliminate options, preferences that weight options, relevant facts about the domain.

Step 3: Format as a Decision Brief

Structure the response as follows:

---

DECISION BRIEF: [Topic]

Question: [Restate the question clearly]

Prior Decisions on This Topic [What was decided before, when, and why — from deepSearch and journalSearch. If nothing found: "No prior decisions found."]

Context Then vs Now [If prior decisions exist: what has changed since then? New constraints, new options, new information.]

Tradeoffs | Option | Pros | Cons | |--------|------|------| | [A] | ... | ... | | [B] | ... | ... |

Recommendation [Clear recommendation with brief rationale. Ground it in the retrieved preferences/constraints. If preferences conflict with the current situation, call it out explicitly.]

Confidence: [High / Medium / Low] — [one sentence explaining why]

---

Rules

• Do NOT answer before completing Steps 1-2. The Brief only has value if it reflects actual memory.
• If memory retrieval returns nothing useful, say so in the Brief — don't fabricate prior context.
• Keep the Brief tight: prior decisions and recommendation are mandatory; tradeoffs table optional for simple choices.
• If the user's question is complex enough to require web research (emerging tech, pricing, etc.), note it in the Recommendation — but always anchor with memory first.