Is alexsjones/llmfit safe?
Yes. alexsjones/llmfit is safe to install. Oathe's behavioral security audit gave the llmfit skill by alexsjones a trust score of 91/100 with 6 findings, none critical or high.
https://github.com/alexsjones/llmfit
Is alexsjones/llmfit safe to install?
llmfit-advisor is a legitimate, well-structured hardware detection skill for recommending locally-runnable LLM models. All monitoring evidence is benign: honeypot canaries are intact, the only network connection during clone was to GitHub, and sensitive file accesses pre-date the clone by six seconds (attributable to the audit framework). The skill's most notable behaviors — modifying openclaw.json and optionally sharing benchmarks to GitHub — are fully declared in SKILL.md and require explicit user action. No prompt injection, hidden instructions, or exfiltration attempts were detected.
What security issues were found in alexsjones/llmfit?
Category Scores
Findings (6)
INFO Sensitive file accesses pre-date clone 0 ▶
Inotifywait recorded accesses to .env, id_rsa, .aws/credentials, .npmrc, .docker/config.json, and gcloud credentials at 15:50:15. The git clone did not begin until 15:50:21. These accesses are attributable to the Oathe audit framework establishing canary baselines, not to any code in the skill.
LOW Git hook present in .githooks/ -8 ▶
.githooks/pre-push (208 bytes, executable) is present in the repository. This hook only activates if a developer explicitly configures git with 'core.hooksPath .githooks'. It will not auto-execute during clone or when the skill is installed. Pre-push hooks run on the developer's machine when pushing code, not on end-user machines.
LOW Installation scripts present but not auto-executed -4 ▶
install.sh and scripts/install-openclaw-skill.sh are present and would download/install binaries if run. Neither is triggered by git clone, npm install, or cargo build. They require explicit user invocation.
INFO bench --share creates GitHub PRs with benchmark data -10 ▶
The llmfit tool's bench --share feature can fork the llmfit GitHub repo, commit benchmark results, and open a pull request. This is significant out-of-band action but is: (a) only triggered explicitly, (b) guarded by a confirmation prompt, (c) previewed via --dry-run, and (d) uses standard GitHub device-flow OAuth. The SKILL.md does not instruct the agent to use this feature automatically.
INFO Skill modifies user agent configuration -5 ▶
SKILL.md instructs the agent to update openclaw.json with Ollama/vLLM model entries and optionally set a default model. This is the skill's stated purpose and is transparently documented. It does not override system-level instructions.
INFO Only expected GitHub HTTPS connection during clone -3 ▶
Network monitoring confirmed a single outbound TCP connection to 140.82.121.4:443 (GitHub's IP) for the shallow clone. DNS resolved via local resolver only. No additional connections were established after clone completed.
Should I install alexsjones/llmfit?
Oathe's verdict for alexsjones/llmfit is SAFE with a trust score of 91/100. Recommendation: Install.