Appearance
Appendix E: One-Page Design Sheet
Use this sheet to sketch a simple agent before implementation.
1. Task
| Item | Answer |
|---|---|
| Goal | |
| Success criteria | |
| Out of scope | |
| Allowed actions | |
| Forbidden actions | |
| Autonomy level |
2. Context
| Section | Content |
|---|---|
| System | |
| Task | |
| Tools | |
| State | |
| History strategy | |
| Latest observation | |
| Token budget |
3. Output Protocol
| Intent | Meaning |
|---|---|
| Tool request | |
| Final answer | |
| Clarification | |
| Abort |
4. Tools
| Tool | Purpose | Risk class | Permissions |
|---|---|---|---|
5. Loop Control
| Control | Limit |
|---|---|
| Maximum iterations | |
| Token budget | |
| Cost budget | |
| Timeout | |
| Retry limit | |
| Stop reasons |
6. Guardrails
| Guardrail | Enforcement point |
|---|---|
| Tool allowlist | |
| Argument validation | |
| Permission scope | |
| Sandbox / isolation | |
| Final-answer check | |
| Audit logging |
7. Observability
| Recorded item | Purpose |
|---|---|
| Run ID | Correlate events |
| Step number | Order loop |
| Model output | Audit raw emission |
| Parsed intent | Audit harness interpretation |
| Tool status | Diagnose action failures |
| Stop reason | Explain termination |
| Tokens/cost | Control spend |