SkillScout
ReviewAudited by ClawScan on May 10, 2026.
Overview
SkillScout is a coherent skill-review catalog, but its strong safety claims appear broader than the provided evidence supports, so users should verify recommendations before installing anything.
Use SkillScout as one input, not as final proof a skill is safe. Before installing a recommended skill, open its review, inspect its source and permissions, verify the catalog/package provenance, and avoid automatic installs based only on the trust score.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
A user or agent may over-trust a 'safe' rating and install or recommend a third-party skill without enough independent review.
These absolute assurances are central to a skill that influences installation decisions. Other provided artifacts describe read-only/no-execution review and an empty blocklist that will be populated later, so the user-facing safety posture appears stronger than the artifact-backed process shown.
Every skill tested. Every skill reviewed. No junk. No malware. ... Final approval by a human reviewer before any skill is listed. No exceptions.
Treat the ratings as advisory, not definitive; verify the specific review, source repository, permissions, and install command before installing a recommended skill.
If the npm package or publisher account is compromised, running the MCP command could execute unexpected code.
Running an unpinned npx command can fetch and execute the latest npm package. This is purpose-aligned for the optional MCP interface, but it is still a supply-chain-sensitive step.
### MCP Server (for agent-to-agent queries) ```bash npx @skillscout/mcp ```
Pin a known version, confirm the npm package publisher/source matches the advertised GitHub project, and inspect the MCP server code before use.
Recommendations could become inaccurate or unsafe if the hosted catalog is stale, altered, or compromised.
The agent's recommendations and trust ordering come from remote JSON. That is expected for a catalog, but stale or compromised catalog data could poison the agent's recommendation context.
curl -s https://nashbot67.github.io/skillscout/data/skills.json | python3 -c ... sorted(matches, key=lambda x: {'safe':0,'caution':1,'avoid':2}.get(x.get('trustScore',''),1))Cross-check important recommendations against the linked source and review files, and do not treat remote catalog entries as executable instructions.