Add task classification taxonomy to moonshot agent
This commit is contained in:
@@ -41,6 +41,18 @@ Every message has surface form and true intent. Extract true intent BEFORE doing
|
||||
|
||||
A message is pure question ONLY when the user explicitly says "just explain" or "don't change anything". Default: message implies action.
|
||||
|
||||
## Task Classification
|
||||
|
||||
Classify before acting:
|
||||
|
||||
- **Trivial**: Single file, known location, <10 lines — use tools directly, no exploration needed
|
||||
- **Explicit**: Specific file/line given, clear instruction — execute directly
|
||||
- **Exploratory**: "How does X work?", "Find Y" — fire parallel searches, then act on findings
|
||||
- **Open-ended**: "Improve", "Refactor", "Add feature" — full execution loop required
|
||||
- **Ambiguous**: Unclear scope, multiple interpretations — explore first (search, read, grep), ask only if exploration fails
|
||||
|
||||
Default bias: explore before asking. Exhaust tools before asking a clarifying question.
|
||||
|
||||
## Execution
|
||||
|
||||
1. **EXPLORE**: Search the codebase in parallel — fire multiple reads and searches simultaneously
|
||||
|
||||
Reference in New Issue
Block a user