multi-llm
Verdict safe, but 2 findings (1 medium, 1 low), mainly unauthorized tool use, social engineering. Most recent scan 2026-02-06.
The multi-llm skill is a LOCAL MODEL SWITCHING utility that intelligently routes tasks to different Ollama models based on task type detection. The skill is SAFE with minor documentation improvements needed. It does NOT exhibit malicious behavior - no data exfiltration, no prompt injection, no credential theft. The network calls are legitimate (localhost:11434 for Ollama API). The skill operates as described: defaults to Claude Opus 4.5 unless 'multi llm' trigger is used, then selects appropriate local models. Primary concerns are documentation completeness (missing allowed-tools, license, compatibility fields) and ensuring graceful degradation when Ollama is unavailable. The skill is well-structured with clear fallback chains and model availability checking.
Scanner's overall assessment (LLM analyzer)
Second scanner: NVIDIA SkillSpector
Independent static run, no language model, over the same files as the Cisco scan above. How the two scanners compare across the whole corpus: Two scanners, one corpus.
What this skill does
Multi-LLM intelligent switching. Use command 'multi llm' to activate local model selection based on task type. Default uses Claude Opus 4.5.
- Package
- 3 files, 18 KB
- SKILL.md
- 751 words, 12+ sections
- Code
- Shell (2)
- Version
- 1.1.0
SKILL.md covers: Multi-LLM - Intelligent Model Switching · What's New in v1.1.0 · Usage · Default Mode (without command) · Multi-Model Mode (with command) · Command Format · Model Mapping · Fallback Chain · Detection Logic · Task Detection Keywords · Examples · Example 1: Coding Task
Findings
MEDIUM Missing allowed-tools Declaration Unauthorized Tool Usellm
The skill does not declare 'allowed-tools' in its YAML manifest. While this field is optional per the agent skills specification, it is recommended to explicitly declare tool usage for transparency. The skill's bash scripts execute shell commands and make network calls (curl to localhost:11434), which should be documented via 'allowed-tools: [Bash]' for clarity.
SKILL.md
allowed-tools: Not specified
Scripts use: curl, bash commands, network calls to localhost:11434LOW Missing License and Compatibility Metadata Social Engineeringllm
The skill is missing 'license' and 'compatibility' fields in its YAML manifest. While these are optional fields, including them improves transparency and helps users understand licensing terms and which environments the skill supports (Claude.ai, Claude Code, API).
SKILL.md
license: Not specified
compatibility: Not specifiedWhat is in the package
| File | Kind | Size |
|---|---|---|
SKILL.md | doc | 5.4 KB |
scripts/fallback-demo.sh | script | 6.5 KB |
scripts/select-model.sh | script | 6.0 KB |
External hosts referenced
github.com ollama.com
Hostnames that appear in the skill's files. Referencing a host is not a finding; compare against the findings above and the skill's stated purpose.
Environment variables referenced
PRIMARY_MODELS DEFAULT_MODEL FALLBACK_MODELS NETWORK_MODEL
Package contents recorded 2026-10-22. The scan above ran against the version current at scan time; the two can differ if the skill has since been updated.
What we checked
- Static analyzer. Pattern rules over SKILL.md and bundled files: known-bad commands, exfiltration URLs, secrets, obfuscation.
- Behavioral analyzer. Checks what the skill asks the agent to do against what its description claims it does.
- Llm analyzer. A language model reads the skill and reports intent-level issues the rules cannot express.
- Trigger analyzer. Looks for instructions that fire on conditions the user did not ask for.
- Meta analyzer. A second model pass that validates, deduplicates and re-scores the other analyzers' findings.
What we did not check
- Runtime behaviour. The skill was not executed; findings come from reading its files.
- Anything fetched at install or run time, including remote scripts, package dependencies, and URLs the skill tells the agent to open.
- Content the analyzers cannot read: images, binaries, and some non-English text.
- Known scanner bypass classes. See the methodology and limitations page.
Sources
- Repository: https://clawhub.ai/skills/mlti-llm-fallback
- Raw scan result: multi-llm-fallback-scan.json
- Badge for your README:
- Request a re-scan