Pre-Flight Check
Benchmarked against: Anthropic — Effort (adaptive complexity) SOP: Cloud UB ub-a0914abf17a7 Scope: All agents, before every non-trivial task
The Pre-Flight Check is a 30-second self-assessment before any task. It scores risk across three dimensions and determines whether to proceed, research first, or escalate.
3D scoring
Each dimension scores 0–2. Total score determines action.
Dimension 1: Blast Radius
| Score | Meaning |
|---|---|
| 0 | Local only (one file, one config) |
| 1 | Module-level (multiple files, one service) |
| 2 | Fleet-wide (cross-ship, cross-agent, production) |
Dimension 2: Knowledge Freshness
| Score | Meaning |
|---|---|
| 0 | Timeless knowledge (design patterns, algorithms) |
| 1 | Perishable but recently verified |
| 2 | Perishable, not recently verified |
Dimension 3: Reversibility
| Score | Meaning |
|---|---|
| 0 | Fully reversible (git revert, config rollback) |
| 1 | Partially reversible (database migration, API change) |
| 2 | Irreversible (data deletion, external publishing) |
Action thresholds
| Total Score | Color | Action |
|---|---|---|
| 0–2 | 🟢 Green | Proceed directly |
| 3–4 | 🟡 Yellow | Search UB first, then proceed |
| 5–6 | 🔴 Red | Search UB + intel_search + report to Captain before acting |
Example
Task: Update LangGraph agent to use new checkpoint API
Blast Radius: 1 (module-level, affects agent pipeline)
Freshness: 2 (LangGraph API is Danger Zone, not recently verified)
Reversibility: 0 (can git revert)
Total: 3 → 🟡 Yellow → search UB first
Integration points
- Tech Freshness: Dimension 2 directly maps to the Danger Zone library list
- HITL Boundary: Score 5-6 always requires Captain visibility (Constitution §5)
- Work Orders: WO submission gates can require Pre-Flight Check record
When to skip
The only time you can skip Pre-Flight Check is for truly trivial tasks:
- Fixing a typo
- Adding a log statement
- Reading files for information
If you're writing code, changing config, or making any decision — run the check.