Safe Operation Patterns
Rule
Classify operations as GREEN (always safe), YELLOW (safe with patterns), or RED (requires human) before executing.
Context
During autonomous operation when selecting and executing any task or action.
Detection
- About to execute any operation autonomously
- Task involves external interaction
- Planning actions with potential impact
- Tool returns empty data (connection OK but no results)
Pattern
Classify before executing, then follow appropriate action:
Classification Table
| Classification | Criteria | Action |
|---|---|---|
| GREEN ✓ | Safe, reversible, autonomous | Execute immediately |
| YELLOW ⚠ | Safe with established pattern | Follow specific pattern (e.g., twitter-best-practices.md) |
| RED ❌ | Financial, irreversible, human judgment | Escalate, don't execute |
Examples:
- Code change → GREEN (reversible)
- Tweet update → YELLOW (has pattern)
- Execute trade → RED (financial)
Empty Data Debugging
When tool connects but returns zero results:
| Check | Finding | Action |
|---|---|---|
curl <source_url> |
Source has data | Tool parsing bug |
curl <source_url> |
Source empty | Genuinely empty, move on |
curl <source_url> |
Different format | Update tool config |
Skip raw inspection if: connection failed, tool reports errors, multiple tools confirm emptiness.
Outcome
- Safe autonomy: No dangerous operations without human
- Efficient progress: GREEN tasks proceed immediately
- Efficient debugging: Distinguish source vs tool bugs quickly
Related
- Autonomous Operation Safety - The "lethal trifecta" pattern
- Escalation vs Autonomy - Decision framework