| name | gosm - Goal-Oriented State Machine |
| description | Goal-Oriented State Machine. Context-adaptive dispatcher that routes to the right analysis depth based on urgency, stakes, and expertise. |
| context | This is the default entry point. When a user needs help thinking through something and hasn't specified a skill, GOSM selects the appropriate depth. |
| output | {"format":"prose"} |
GOSM - Goal-Oriented State Machine
Input: $ARGUMENTS
Principle
Match analysis depth to the situation. Don't overthink low-stakes decisions. Don't underthink high-stakes ones.
Context Assessment
Assess these five factors to select the right variant:
| Factor | Options |
|---|
| Urgency | URGENT (minutes) / SOON (hours) / NORMAL (days+) |
| Stakes | HIGH (major impact) / MEDIUM / LOW (minor) |
| Expertise | EXPERT (user knows domain) / INTERMEDIATE / NOVICE |
| Action Cost | CHEAP (reversible) / EXPENSIVE (one-shot) |
| Information | RICH (lots known) / MODERATE / SPARSE |
Variant Selection
| Context | Variant | What It Does |
|---|
| URGENT + any | Lite | Find the ONE critical assumption. Test it. Act. |
| LOW stakes + CHEAP | Quick | Sanity check. If nothing's obviously wrong, proceed. |
| EXPERT + HIGH confidence | Check | Validate the user's plan. Find blind spots. |
| Post-action / reflecting | After | Learn from what happened. Update model. |
| MEDIUM stakes + NORMAL time | Standard | Balanced analysis: key claims, brief ARAW, action. |
| HIGH stakes + EXPENSIVE + NOVICE | Full | Comprehensive: invoke /pce |
| EXPLORING + NO DECISION NEEDED | Explore | Map the space, take a position, surface tensions. Not everything is a decision. |
User can override: "quick", "full", "just check this", "reflect on what happened", "just thinking about this".
Lite (< 5 minutes)
For urgent situations. Find the ONE assumption that matters most.
- What's the core claim? (one sentence)
- What's the critical assumption? (the one thing that changes everything if wrong)
- Quick ASSUME WRONG: If that assumption is wrong, what's the best alternative? What's the risk?
- Action: What to do right now. How to verify it worked.
- Revisit when: Time permits / new info arrives / action didn't work.
Quick (< 2 minutes)
For low-stakes, reversible decisions. Don't overthink.
- Sanity check: Any obvious constraints violated? Any clearly better option being ignored? Will I regret not thinking harder?
- If all clear → Proceed. Note the undo condition.
- If anything flagged → Decide if it's worth deeper analysis.
Check (< 10 minutes)
For experts who have a plan. Don't explore — validate.
- Blind spot scan: Obvious failure modes? Missing stakeholders? Unrealistic assumptions? Resource gaps? Timeline issues?
- Strongest argument against: What's the best case for NOT doing this?
- Verdict: PROCEED / PROCEED WITH CAUTION (issues noted) / PAUSE (significant issue) / RETHINK (fundamental problem)
After (< 15 minutes)
For reflecting on actions already taken.
- Expected vs actual: What did you think would happen? What did happen? What's the gap?
- What this reveals: For each learning — what to do differently.
- Updated model: What did you believe before? What do you believe now?
- Next action: Based on what was learned.
Standard (15-30 minutes)
Balanced analysis for medium-stakes situations.
- Classify: GOAL / PROBLEM / QUESTION / DECISION / SITUATION
- Key claims: Surface + hidden assumptions (2-4 claims)
- ARAW the most important claim: Assume Right → what follows? Assume Wrong → what alternatives?
- Goal journey: Current state → desired state → what serves this?
- Contrarian view: What's the strongest challenge?
- Actionable filter: What can the user actually do?
- Crux question: The one question that matters most.
- Recommended action: Specific, with verification.
Explore (5-15 minutes)
For thinking, wondering, or exploring without a decision to make.
- What's the interesting question here? Not the stated question — the one underneath it.
- Map the key distinctions. What concepts are being conflated? Where are the boundaries?
- Take a position. Exploring doesn't mean refusing to commit. "I think X because Y, but Z is the strongest counter."
- Surface the tension. What contradiction or surprise emerges from thinking carefully about this?
- Open threads. What's worth exploring further? Name them, don't pursue all of them.
Full (30-60+ minutes)
For high-stakes, expensive, novel situations. Invokes the full procedure engine.
→ /pce $ARGUMENTS
Pre-Completion Check
