thoughtbox:decision-38be03— agent skill

Use Thoughtbox branching to evaluate: $ARGUMENTS

Phase 1: Frame the Decision

1. Create a session titled Decision: [topic].
2. Record thought 1: state the decision clearly.

• What are we choosing between?
• What constraints apply?
• What does success look like?

1. Identify 2-4 distinct options (if not already provided by the user).

Phase 2: Branch and Explore

For each option, create a branch from thought 1 using branchFromThought and branchId. Within each branch, gather evidence, assess pros/cons, and identify risks using reasoning thoughts.

async () => {
  await tb.thought({
    thought: "Option A: PostgreSQL — ACID compliance, mature ecosystem, strong JSON support via jsonb",
    thoughtType: "reasoning",
    nextThoughtNeeded: true,
    thoughtNumber: 2,
    totalThoughts: 15,
    branchFromThought: 1,
    branchId: "postgresql"
  });
}

Each branch should contain at minimum:

• Evidence: concrete facts supporting or undermining the option
• Pros/cons: structured assessment against the stated constraints
• Risks: what could go wrong, how recoverable is it
• Dependencies: what else changes if we pick this option

Use external tools (Grep, Read, web search) to gather real evidence rather than reasoning from memory alone.

Phase 3: Cross-Verify

Compare branches pairwise:

1. Where do options agree? (shared strengths indicate low-risk ground)
2. Where do they contradict? (genuine tradeoffs to weigh)
3. Which criteria does each option win on?
4. Are any options strictly dominated? (worse on every axis — eliminate)

Record cross-verification as a reasoning thought on the main trunk (no branchId), referencing the branch thoughts by number.

See thoughtbox://guidance/parallel-verification for the parallel exploration pattern.

Phase 4: Converge

Record a decision_frame thought with confidence and options. Exactly one option must be selected: true.

async () => {
  await tb.thought({
    thought: "PostgreSQL wins: ACID required for financial data, jsonb handles semi-structured needs, team has existing expertise",
    thoughtType: "decision_frame",
    confidence: "high",
    options: [
      { label: "PostgreSQL", selected: true, reason: "ACID + jsonb + team expertise" },
      { label: "MongoDB", selected: false, reason: "Better schema flexibility but ACID trade-off unacceptable" },
      { label: "DynamoDB", selected: false, reason: "Vendor lock-in, team unfamiliar" }
    ],
    nextThoughtNeeded: false,
    thoughtNumber: 15,
    totalThoughts: 15
  });
}

State the rationale in one clear sentence. If confidence is low or medium, explain what additional information would raise it.

Phase 5: Persist

1. Create a knowledge entity for the decision:

   async () => {
     await tb.knowledgeGraph({
       operation: "create_entities",
       entities: [{
         name: "Decision: [topic]",
         entityType: "Insight",
         observations: [
           "Chose [option] because [rationale]",
           "Rejected [option] because [reason]",
           "Key constraint: [constraint that drove the decision]"
         ]
       }]
     });
   }

1. Create relations to relevant existing entities (components, services,

specs) that the decision affects.

1. Add an observation with the full rationale so future sessions can

recover the reasoning without replaying the branch exploration.

When to Bridge to an ADR

If the decision is architectural — meaning it constrains future implementation choices, crosses module boundaries, or is expensive to reverse — bridge to a formal Architecture Decision Record:

1. Run the hdd skill to stage an ADR in .adr/staging/.
2. Reference the Thoughtbox session ID in the ADR's context section.
3. The decision_frame thought becomes the ADR's decision; the branch

exploration becomes the ADR's considered alternatives.

Non-architectural decisions (library choice within a module, naming conventions, test strategy for a single feature) stay in the knowledge graph without an ADR.