agentleFS
GuilscGuilsc/modelcompass/skills/model-selector/SKILL.md

model-selector

Skill1 starsChanged 27 days agoMITView on GitHub

Recommend context-appropriate AI models, reasoning effort, and cost-quality tradeoffs for human or agent tasks. Use when the user asks which model to choose, compares settings, or explicitly requests model selection guidance. Respect selected models and ask before switching.

3 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions3 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 3

    1. cclank/lanshu-awesome-ai-video-kit★ 418same name
    2. bordenet/superpowers-plus★ 8same name
    3. genfeedai/genfeed.ai★ 4same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Prefer one capable lead model across tightly coupled phases when handoffs waste context.

    line 16
  2. Distinctive:

    Never claim a switch was made unless the host supports it and confirms it.

    line 18
  3. Very specific:

    If the host cannot resolve references/, package those files alongside this skill in an installable directory and adjust paths.

    line 44

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Model Selector

Select models by whole-task outcome, not individual microsteps or cheapest token. This skill is advisory: it does not imply you can actually switch the underlying model.

Procedure

  1. Identify the full workflow, required output modality, quality floor, privacy restrictions, available platform/models, budget, session continuity, and whether the user has already selected a model.
  2. Apply hard eligibility gates: native generation vs understanding, tool availability, language/format fidelity, context capacity, safe operation, and authorized spending. Exclude unqualified candidates before ranking.
  3. Read only relevant modules from references/: task-matrix.md for a matched task; model-catalog.md for model capabilities; reasoning-levels.md for effort; cost-efficiency.md for sessions and cost; sources.md to check freshness. Use data/ only for a machine-readable shortlist, never assume it is exhaustive. For benchmark-informed comparisons, read references/benchmarks.md and inspect data/benchmarks.json (optionally via python scripts/benchmarks.py from this skill directory). Cite source-specific metric/version/settings and dates; preserve missing data and estimated evidence. Never combine scores across sources or treat an aggregate as independent corroboration.
  4. Present Best Fit, Best Value, Budget, Fallback as justified candidates with evidence tags: D provider documented, E editorial assessment, T independently tested, U unverified. Do not invent rankings, tokens, API IDs, prices, or benchmark scores.
  5. Prefer one capable lead model across tightly coupled phases when handoffs waste context. Use deterministic utilities for inventory, hashes, moves, builds, tests and validation where possible. Escalate reasoning effort only for concrete unresolved difficulty; do not assume maximum effort is optimal.
  6. If the user explicitly specified a capable model or effort, respect it. In default Advisory mode, suggest an alternative only if materially beneficial, at most once for the same circumstances. Ask whether to use the recommendation, keep their choice, or enter Grill Me comparison mode. Strict preference suppresses cost nudges. Auto-select requires explicit delegation and must obey scope and constraints.
  7. Never claim a switch was made unless the host supports it and confirms it. Never delete, move, transmit sensitive data, or incur costs without appropriate authorization. When user rejects an alternative, stop pressing unless facts materially change.

Existing selectors and routers

If the host already uses a model selector (/model, platform picker) or an intelligent router (for example, OmniRouter / OmniRoute, OpenRouter Auto Router, LiteLLM, RouteLLM or Not Diamond), operate as an advisory layer and preserve its execution authority. Prefer Observe, Advise, or Defer coexistence modes; these are policies, not API integrations. Check actual selected-model metadata where supported, and do not confuse provider fallback with model selection. Do not install competing automatic policy hooks, silently change routing tables, or claim to have selected a provider when the router may pick or fall back elsewhere. If the user has delegated model selection to the router, do not second-guess every request; offer a second opinion only when explicitly requested or when a meaningful quality/cost issue is evident. Factor in prompt-cache resets and context-transfer costs before suggesting a mid-session switch.

Response format

Task / quality floor: ... Current choice (if any): ... Best Fit / Best Value / Budget / Fallback: ... Why and caveats: quality, reliability, total cost (including switching, retries and context), modalities, evidence/confidence, price freshness. Action: proceed with user preference, suggest a single switch, or ask only an essential missing constraint.

References

  • references/decision-rules.md: gates, decision tree, scoring and selection policy.
  • references/task-matrix.md: initial context-to-model candidates.
  • references/model-catalog.md: providers and capability notes.
  • references/reasoning-levels.md: effort guidance.
  • references/cost-efficiency.md: complete-outcome efficiency.
  • references/advisory-protocol.md: user override, Grill Me, strict mode.
  • references/sources.md: source URLs and maintenance requirements.
  • references/benchmarks.md: benchmark evidence, freshness, missing data and update policy.
  • references/router-interoperability.md: coexistence with model pickers, gateways and intelligent routers.

If the host cannot resolve references/, package those files alongside this skill in an installable directory and adjust paths. Full portable specification: MODEL_HELPER.md.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.