llama-cpp

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

Quality

81%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

76%Weight 40%Scale 1-5

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

A highly actionable, well-structured reference with executable commands and properly signaled one-level-deep bundle references. The main gap is workflow clarity: setup and batch paths lack explicit validation/feedback checkpoints.

Suggestions

Add a validation step after model download (e.g., verify the GGUF loads with a smoke-test inference) and a verify step after starting llama-server (e.g., curl the /health or /v1/models endpoint) so the Quick start has explicit checkpoints.

For the batch-processing pattern, add a feedback loop: check output validity / count completed prompts and re-run failures, since batch operations should not score above 3 without validation per the rubric cap.

Move the benchmark and supported-models tables into a reference file (e.g., references/benchmarks.md) to reduce inline reference-grade content and tighten progressive disclosure toward a 5.

Dimension	Reasoning	Score
Conciseness	Mostly lean with executable snippets and compact tables, but minor repetition (e.g., Q4_K_M recommended in both the format table and the 'Choosing quantization' block) and a few restatable comments could be trimmed.	4 / 5
Actionability	Copy-paste-ready commands cover the common cases end-to-end — install/build per backend, download GGUF, run chat, start OpenAI-compatible server, batch processing, grammar-constrained JSON, and context sizing.	5 / 5
Workflow Clarity	The Quick start lays out a clear install → download → run sequence, but there are no validation/verification checkpoints, and the batch-processing pattern operates without a feedback loop; the batch-operations cap holds this at 3.	3 / 5
Progressive Disclosure	Well-signaled one-level-deep references to real bundle files (quantization.md, server.md, optimization.md, all present) with organized sections, though a fair amount of reference-grade material (benchmark tables, supported-models list) remains inline.	4 / 5
	Total	16 / 20 Passed

Description

87%Weight 40%Scale 1-5

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 pairs a concrete capability statement with an explicit 'Use for...' trigger clause and a well-scoped non-NVIDIA niche. Minor room to add a few more synonyms/file extensions for trigger terms.

Dimension	Reasoning	Score
Specificity	Lists several concrete capabilities ('Runs LLM inference on CPU, Apple Silicon, and consumer GPUs', 'Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU') but coverage of actions is slightly narrower than the comprehensive 5-anchor example.	4 / 5
Completeness	Clearly answers 'what' (runs LLM inference with GGUF quantization for reduced memory and speedup) and explicitly answers 'when' via the 'Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable' trigger clause.	5 / 5
Trigger Term Quality	Strong natural trigger phrases ('edge deployment', 'M1/M2/M3 Macs', 'AMD/Intel GPUs', 'when CUDA is unavailable') that users would actually say, with a few synonyms/extensions (e.g., '.gguf', 'Raspberry Pi') missing.	4 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (non-NVIDIA / CPU / Apple Silicon inference) with distinct triggers ('without NVIDIA hardware', 'when CUDA is unavailable') that minimize overlap with CUDA-first skills like TensorRT-LLM or vLLM.	5 / 5
	Total	18 / 20 Passed

Validation

93%

Warnings & errors only

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

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-inference/llama-cpp/SKILL.md
Commit: fcd2e11

Reviewed: 2 days ago

Table of Contents

Discovery Implementation Validation

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.