| name | mem-load |
| description | Auto-load domain knowledge at task start. Reads /knowledge/INDEX.md, finds relevant domain folders for the current task, loads rules.md (apply by default) and hypotheses.md (check if testable). Invoke manually when switching domains mid-conversation or when working in a domain you haven't loaded yet. |
Memory Load
Load relevant domain knowledge before starting work. This ensures confirmed rules are applied and testable hypotheses are tracked.
Steps
- Read the index — read
knowledge/INDEX.md to see all available domains
- Match domains — identify which domains are relevant to the current task based on the user's request, file paths, and topic keywords
- Load rules — read
rules.md in each relevant domain. These are confirmed patterns (5+ confirmations). Apply them by default without being asked
- Load hypotheses — read
hypotheses.md in each relevant domain. Note which ones could be tested or observed during this task
- Brief the user — one-line summary:
Loaded: [domain1] (N rules, M hypotheses), [domain2] (N rules, M hypotheses)
Watching: [hypothesis that could be tested today]
What to Do With Loaded Knowledge
- Rules: Follow them. They're confirmed patterns. If a rule conflicts with the current task, note the conflict but still follow the rule unless the user explicitly overrides
- Hypotheses: Watch for confirmation or contradiction during the task. Don't change your approach based on hypotheses — they're unconfirmed. Just observe and note for
/mem-save later
- Knowledge: Use as context. These are facts that help you understand the codebase
Load multiple domains when tasks span boundaries. When uncertain, load the domain — better to have context you don't need than miss a rule you should follow.
