CtrlK
BlogDocsLog inGet started
Tessl Logo

quant-recipe-search

Use when the user asks to find, search for, or optimize the best quantization recipe for a model, including direct requests like "find the best quantization recipe and generate a PTQ checkpoint." Guides the multi-candidate loop: choose compute-vs-memory success metrics, select ModelOpt recipe baselines, design AutoQuant/manual recipe deltas, interpret sensitivity, and decide next candidates. Do NOT use for a single known PTQ recipe run (use ptq), serving (use deployment), creating/running evals (use evaluation or launching-evals), monitoring jobs (use monitor), MLflow browsing (use accessing-mlflow), or comparing completed baseline-vs-candidate scores only (use compare-results).

80

Quality

100%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

100%Weight 40%Scale 1-3

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured strategy skill body that is concise, concrete, and actionable for a complex iterative search, with clearly sequenced workflows, explicit gating/validation checkpoints, and clean one-level-deep reference files.

DimensionReasoningScore

Conciseness

The body assumes Claude's competence with no preamble explaining known concepts (e.g., what quantization or AutoQuant is); every section carries domain-specific decision content with no fluff or padding, even though it is long for a genuinely complex multi-axis search task.

3 / 3

Actionability

As an instruction-only strategy skill, it provides concrete and specific guidance: named axes (e.g., "Numeric format: FP8/W8A8, NVFP4/W4A4"), named module families, explicit decision branches, named sibling skills to delegate to, and a defined portfolio-table schema; the absence of code is not penalized under the instruction-only carve-out because the guidance is actionable.

3 / 3

Workflow Clarity

Two clearly sequenced multi-step processes (Design Workflow and Iteration Loop) include explicit validation checkpoints such as the "Gate before scaling" step, rerun-on-noise guidance, and promote-only-when-validated feedback loops.

3 / 3

Progressive Disclosure

SKILL.md serves as an overview with two well-signaled, one-level-deep references (recipe_iteration.md for design/search-space/sensitivity details, and qwen36_case_study.md conditionally for Qwen3.5/3.6 specifics); both referenced files exist under ./references/, and content is appropriately split with easy navigation.

3 / 3

Total

12

/

12

Passed

Description

100%Weight 40%Scale 1-3

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, third-person description that concisely states concrete capabilities, natural trigger terms, explicit when-guidance, and clear disambiguation from sibling skills via a Do-NOT-use clause.

DimensionReasoningScore

Specificity

Lists multiple concrete actions ("choose compute-vs-memory success metrics," "select ModelOpt recipe baselines," "design AutoQuant/manual recipe deltas," "interpret sensitivity," "decide next candidates"), matching the score-3 anchor for listing multiple specific concrete actions.

3 / 3

Completeness

Explicitly answers both what ("Guides the multi-candidate loop: choose..., select..., design..., interpret..., decide next candidates") and when ("Use when the user asks to find, search for, or optimize..."), with an explicit Use-when trigger clause.

3 / 3

Trigger Term Quality

Includes natural terms users would say ("find, search for, or optimize the best quantization recipe") plus a quoted direct request ("find the best quantization recipe and generate a PTQ checkpoint"), giving good coverage of common phrasings.

3 / 3

Distinctiveness Conflict Risk

A clear niche is sharpened by the explicit negative-guidance clause naming sibling skills (ptq, deployment, evaluation, launching-evals, monitor, accessing-mlflow, compare-results), making it unlikely to trigger for the wrong skill.

3 / 3

Total

12

/

12

Passed

Validation

100%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository
NVIDIA/Model-Optimizer
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.