This table lists when to consider each preset, never which is better. Better depends on your work, not on the preset.
| Your situation | Consider | Reason (from the preset definitions, not my judgment) |
|---|---|---|
| Unsure, first time | Standard | The widest tool surface: file editing, shell, search, Skills, plans, subagents, workflows |
| Multi-step work expressible as one program | PTC | Tools presented via an SDK so the model composes one TS program; round trips can collapse into one |
| Only run commands and edit files | Minimal | Two tools only: persistent shell and str_replace_editor; no Skills, plans, or subagents |
| Authoring or debugging a custom preset | Creator | Standard capability plus runtime inspection, plugin experiments, and authoring guidance |
| Long session with many follow-ups | Not minimal | minimal/agent.cordis.yml states Context compaction is absent |
| No usable code runtime | Not PTC | PTC requires a codeRuntime whose language has a registered SDK renderer, or prompt assembly fails |
Two hard constraints worth their own line
- Minimal means no compaction. Not 'slightly fewer features' — long sessions hit the ceiling. Use it for short, few-turn tasks.
- PTC needs a code runtime, and not just any: the language must have a registered SDK renderer (TypeScript or Python). Missing means a failed start, not degraded operation.
Last rule of thumb: when unsure, use standard. It is not optimal, but it is the one where 'other people already hit this' is easiest to find — valuable when troubleshooting.