Decide

/decide

One skill. Three modes. Every decision produces a reasoning artifact with a priority level.

Setup

The vault path is configured during memento init. If not set, ask the user where their vault lives.

Mode Detection

Mode: --research (Ingest First)

Extract knowledge from the source before deciding.

Extraction Engine (5 Fields)

1. Core argument — what it argues, not what it's about (1-2 sentences)
2. Why it holds — evidence, reasoning, mechanism (3 bullets max)
3. Technical specifics — numbers, code patterns, tools. "None" if none
4. Execution pattern — the craft move worth stealing. Omit if N/A
5. Open questions — what connects to existing work? What assumption might break?

After extraction, ask: "Does this inform a decision, or is it reference material?"

• Decision → continue to OODA loop
• Reference only → save to project Research/ folder, stop

Mode: --reframe

Force a perspective shift before presenting options:

1. Invert the question — "What if we did the opposite?"
2. Change the timeframe — "What matters in 6 months vs today?"
3. Change the stakeholder — "What would the end user prioritize?"
4. Expose assumptions — "What are we assuming that might not be true?"

Then enter the OODA loop with the expanded frame.

Retrieval Gate

Before presenting options, search the vault for prior decisions on this topic:

1. Scan project _context.md Active Reasoning Artifacts table for [D] entries matching the decision domain
2. Glob {vault_path}/Knowledge/*{topic}* for existing knowledge notes
3. Glob {vault_path}/Projects/*/Decisions/*{topic}* for cross-project decisions
4. If prior decision found:
   • Surface it: "Existing decision: [D] {statement} — {date}"
   • Ask: "Reaffirm, revise, or override?"
5. If no prior decision: proceed with fresh analysis

OODA Decision Loop (All Modes)

Observe

• Read current project _context.md and NEXT.md
• Check existing decisions and Active Reasoning Artifacts table (avoid re-deciding settled questions)

Orient

Present the decision landscape:

• Options (2-5): each with concrete pros, cons, and effort estimate
• Recommendation: which option and why
• Confidence: High (>85%) / Medium (60-85%) / Low (<60%) — show reasoning

Decide

Ask the user to choose. Accept their decision even if it differs from recommendation.

Act — Record the Reasoning Artifact

Step 1: Assess Priority

Claude proposes confidence: High / Medium / Low

Then ask the user: "High or low impact?"

Derive priority from the matrix:

If priority is noise: ask "Store it anyway or discard?" Default: discard.

Step 2: Record

Inline format:

`[D] <conclusion> — invalidates if <trigger> [priority] [YYYY-MM-DD]`

If the decision is "not now, but later when X" → plant a seed instead:

`[S] <idea> — activates when <condition> [priority] [YYYY-MM-DD]`

Seeds are checked at session-start. When the activation condition is met, the seed surfaces and becomes a /decide candidate.

Inline only (settled or noise): Add to _context.md Active Reasoning Artifacts table. Full artifact file (critical or volatile): Create in Decisions/ folder + inline entry.

Significance Heuristic