Manual Instinct Creation
Overview
Save patterns and insights from the current session as instincts. This skill bridges the gap between automatic instinct detection (arc-observing) and structured reflection (arc-reflecting) — it handles ad-hoc "I want to remember this" moments.
Quick Reference
| Task | Command |
|---|---|
| Save instinct | node "${SKILL_ROOT}/scripts/recall.js" save --id {id} --trigger "..." --action "..." --domain {d} --project {p} |
| Check duplicate | node "${SKILL_ROOT}/scripts/recall.js" check-duplicate --id {id} --project {p} |
Workflow
- •Receive user's natural language description of the pattern
- •Infer structured fields from the description:
- •
id: kebab-case identifier (e.g.,always-run-tests-first) - •
trigger: When does this apply? (e.g., "when starting a new feature") - •
action: What to do? (e.g., "run existing tests before making changes") - •
domain: Category (e.g.,testing,debugging,workflow) - •
evidence: Supporting context from the session
- •
- •Preview the complete instinct for user confirmation
- •Check duplicate before saving
- •Save via instinct-writer with:
- •
source: 'manual' - •
confidence: 0.50(starting confidence for manual instincts) - •
maxConfidence: 0.90(manual instincts use full MAX_CONFIDENCE)
- •
When to Use
- •User explicitly says "/recall" or "remember this"
- •User identifies a reusable technique during work
- •User wants to preserve a pattern without waiting for automatic detection
- •User wants to save an insight from the current conversation
When NOT to Use
- •Pattern was already auto-detected by arc-observing (use confirm instead)
- •User wants to analyze multiple diaries for patterns (use /reflect)
- •User wants to cluster instincts into higher-level skills (use /learn)
- •User is just discussing patterns without wanting to save them
Key Principles
- •User confirmation required: Always show preview before saving
- •Single instinct per invocation: Don't batch-create multiple instincts
- •Duplicate awareness: Always check-duplicate before saving
- •Accurate inference: Ask clarifying questions if trigger/action are ambiguous