assess-model-fit
Assess model fit for task.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 4
- Forks
- 1
01 前言 Front Matter
- name
- assess-model-fit
- description
- Assess model fit for task.
- allowed-tools
- [ "Bash", "Read", "Edit", "Write" ]
- user-invocable
- true
02 资源文件 1
- 脚本scripts/assess-model-fit.sh8.5 KB
03 SKILL.md 正文
assess-model-fit — Evaluate current model capability
Determine whether the current Claude model is sufficient for your task, or recommend escalation to a higher-level model. Runs in procedural mode (checklist guidance) or executable mode (analyze a task and auto-recommend).
Procedure
Assess model fit (manual, procedural mode)
-
Identify the current model. Do not guess it: run
python3 scripts/check-session-runtime.py, which reads the session's own transcript (seeverify-running-model). Common tiers:- Tier 1: Fable 5 (
claude-fable-5), Opus 4.8 (claude-opus-4-8) - Tier 2: Sonnet 4.6 / 5 (
claude-sonnet-4-6) - Tier 3: Haiku 4.5 (
claude-haiku-4-5-20251001)
- Tier 1: Fable 5 (
-
Score your task against these criteria. A task needs escalation or orchestrator-worker delegation if it hits any of these:
- Deep multi-step reasoning: more than 5 logical steps, complex dependencies, or architectural design decisions
- Code review rigor: assessing code for subtle bugs, security gaps, performance, or architectural issues (not just syntax)
- Large context window needed: task involves many files, long documents, or substantial history to reason over
- Complex decomposition: breaking down an ambiguous problem into sub-tasks and choosing the right approach (not following a clear spec)
- Uncertain scope: task requirements are vague and need clarification by reasoning
- Novel problem: no standard solution applies; requires creative or exploratory thinking
-
Determine Orchestration Strategy:
- Solo Executor (Tier 2/3): The task is straightforward, bounded, well-specified, and single-purpose.
- Orchestrator + Worker Delegation (Tier 1 Conductor + Tier 2 Workers): The task is large, multi-stage, or complex. Use Tier 1 to plan and decompose into tasks with strict stop conditions, spawn Tier 2 workers to implement, and audit before merging.
- Escalate Model: If the current model struggles with reasoning or loops on a bug, escalate immediately to Tier 1.
-
If escalation needed, invoke
/select-modelto determine the target. Describe your task, andselect-modelwill recommend the optimal tier and suggest a config update if desired.
Executable mode (auto-analysis and auto-chaining)
Instead of running the checklist manually, invoke the skill with a task description.
The script will analyze the task, output an assessment, and if escalation is needed,
automatically call select-model:
/assess-model-fit --task "I need to refactor a large REST API module and add comprehensive unit tests"
The script reads your current model, evaluates the task complexity, and either gives
you a go-ahead or chains into select-model with your task details.
How to use
- Procedural mode (manual checklist): Read this procedure, run through the scoring criteria, decide if escalation is needed.
- Executable mode (auto-analysis): Invoke
/assess-model-fit --task "<your task description>"and let the script recommend a verdict and model. - When in doubt: Use executable mode — it's faster and catches nuance the manual checklist might miss.
- Broader stack recommendation (recm): Use
recmwhen deciding across harnesses, billing tiers, or multi-engine review rather than within-session Claude escalation.
04 同仓库更多 Skills
wrap-up
End-of-session verification and UMS.
workaround-watcher
Watch upstream issue for fix.
wn
→ whats-next.
whats-next
Recommend concrete next step or action.
wait-for-results
Wait for pipeline/CI results.
wai
→ send-notes.