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

72%

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-organized guide with clean progressive disclosure to real reference files. Its weaknesses are length that could be trimmed and missing validation/feedback checkpoints in the run and batch workflows.

Suggestions

Add a verification step after install/download (e.g., a quick "./llama-cli -m model.gguf -p 'hi' -n 8" smoke test) and a validation checkpoint for batch processing so the workflow clears the level-3 bar.

Trim the performance benchmark/cost tables and the supported-models list, or move them into a reference file, to reduce token load without losing the actionable quick-start content.

Remove the duplicated GitHub/HuggingFace links from the Resources section since equivalent links already appear earlier in the body.

Dimension	Reasoning	Score
Conciseness	The ~250-line body is mostly efficient and free of basic-concept padding, but benchmark/cost tables, the supported-models list, and a Resources section that repeats the GitHub/HuggingFace links already cited could be tightened; it is not the lean 'every token earns its place' level.	2 / 3
Actionability	Provides fully executable, copy-paste-ready commands throughout — "brew install llama.cpp", "make LLAMA_METAL=1", "huggingface-cli download", "./llama-cli", "./llama-server", and a curl client request — with concrete flags and examples.	3 / 3
Workflow Clarity	The install -> download model -> run -> server sequence is clear, but there are no explicit validation checkpoints or feedback loops, and the batch-processing example ("cat prompts.txt \| ./llama-cli ...") lacks verification, which caps workflow clarity at 2 per the rubric guidelines.	2 / 3
Progressive Disclosure	The body is an overview with a well-signaled References section pointing one level deep to three real bundle files (quantization.md, server.md, optimization.md), each labeled with a short description, matching the clear-overview-with-one-level-deep-references anchor.	3 / 3
	Total	10 / 12 Passed

Description

90%

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 with an explicit trigger clause and distinct non-NVIDIA niche. The only minor weakness is specificity, since the action set is limited to running inference and quantization rather than a broad list of operations.

Dimension	Reasoning	Score
Specificity	Names the domain and concrete actions like "Runs LLM inference" and "Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10x speedup", but does not enumerate a broad list of distinct operations, matching the 'names domain and some actions, but not comprehensive' anchor rather than the multi-action level-3 example.	2 / 3
Completeness	Explicitly states what it does ("Runs LLM inference... Supports GGUF quantization...") and when to use it via an explicit "Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable" trigger clause.	3 / 3
Trigger Term Quality	Includes natural terms users would say — "CPU", "Apple Silicon", "M1/M2/M3 Macs", "AMD/Intel GPUs", "when CUDA is unavailable", "edge deployment" — giving good coverage of the hardware scenarios a user would naturally mention.	3 / 3
Distinctiveness Conflict Risk	Carves a clear niche — non-NVIDIA LLM inference on CPU/Apple Silicon/AMD/Intel for edge deployment — with distinct triggers ("without NVIDIA hardware", "when CUDA is unavailable") unlikely to conflict with CUDA-first tools.	3 / 3
	Total	11 / 12 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: 52845c3

Reviewed: about 13 hours 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.