other-model-effort
Skill0 starsChanged 9 days agoMITView on GitHub
Use when the running model can start and follow subagents (any other model skips this skill) and the user types /other-model-effort, or it is about to start an agent, a subagent or a run on a model that is not Claude, in any tool, and its model or reasoning effort must be picked, when such a run seems to need a stronger or a cheaper model, or when a task with several agents is about to go to a model that is not Claude.
In this repository12 other files it gives its agents.
- chrome-bidiSkill
- claude-model-effortSkill
- clean-codeSkill
- clean-commitsSkill
- drive-projectSkill
- limited-runSkill
- progress-boardSkill
- resource-guardSkill
- stop-resume-agentsSkill
- strip-ai-tracesSkill
- using-git-worktreesSkill
- writing-skillsSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Only a model that can start subagents and follow them uses this skill.
- Distinctive: line 14
Never pick a model or an effort for these tools from habit or from the tool's default: the row decides.
- Distinctive: line 18
Every provider sells a light, an all-round and a top model.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Picking a model and an effort level outside Claude
Only a model that can start subagents and follow them uses this skill. A model that cannot (its tool has no subagent or task feature, or it loses track of tool calls over a long run) ignores this skill entirely and does its own task alone.
The task tables live in one place, the claude-model-effort skill (../claude-model-effort/SKILL.md, next to
this folder). Read the row for the task there, then translate its Claude names into the three tiers below.
Never pick a model or an effort for these tools from habit or from the tool's default: the row decides.
Tiers
Every provider sells a light, an all-round and a top model. Price and speed tell them apart, never a name.
| Claude name in the table | Tier | How to spot it in any provider's list |
| Haiku | light | the cheapest and fastest of the provider's list |
| Sonnet | all-round | the middle of the provider's price list, even when the tool starts on another model by default |
| Opus | top | the provider's strongest reasoning model, the most expensive of the list |
A provider with two tiers only: a missing light tier becomes all-round, a missing top tier becomes all-round one effort step higher. A local model runs as light, whatever its name, unless the user says otherwise.
Effort
The table's effort words map onto whatever the tool calls its reasoning setting (effort, thinking level, reasoning budget).
| Table | In the tool |
| low | the lowest setting that still reasons |
| medium | the middle setting of the tool's list, whatever its default is |
| high | one step above medium |
| xhigh | the highest setting below max, unlimited or unbounded |
| max | the tool's max, unlimited or unbounded setting |
Unlike Claude, these models may run at max on any task: their highest setting costs little. The table's effort is where a run starts, and any run may go up to max. The tier still stays between the minimum and the ceiling. A tool or model with no reasoning setting (often the light tier) takes the tier alone and drops the effort.
Several agents
A task that needs several agents at once (a long agentic task, a large feature split across agents, a workflow) never goes to a model that cannot run them: it goes to a model that can, or to Claude.
Rules
Rules 2, 4, 5 and 6 of claude-model-effort hold here: effort sets depth and not quality, the user's choice wins, the highest minimum wins for mixed tasks, two failures move up one step (effort first, up to max, then the tier, never past the ceiling's tier). Rule 1 does not hold: max is allowed. Rule 3 becomes: set the model and the effort where the tool takes them (its command-line flags, its config file, the agent definition), on the run itself, never as a global default the user did not ask for.
/other-model-effort <task>
Find the row in claude-model-effort as its /claude-model-effort <task> part says, then answer in the
user's language with the minimum and the ceiling written as tier and effort (all-round medium, top
high), and the reason, in two or three lines. Never name a model of any provider: the
tier is the answer. When the task needs several agents, say whether that model can
run them.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.