oathe / scan / reports / caution
Is skills/find-skills safe?
Use with caution. skills/find-skills is not clearly unsafe, but the audit flagged risks to review before you install it. Oathe's behavioral security audit gave the find-skills skill by skills a trust score of 62/100 with 6 findings (2 high).
https://github.com/vercel-labs/skills/tree/main/skills/find-skills
Is skills/find-skills safe to install?
The find-skills skill contains no direct malicious content or prompt injection, but its design creates significant architectural risk: it acts as the trusted gateway to all other skill installations, instructs the agent to consult an uncontrolled external domain (skills.sh) for recommendations, and directs silent global installation using -g -y flags that eliminate user consent. A compromised skills registry or manipulated leaderboard would allow an attacker to leverage this skill's trusted position to silently install malicious agents system-wide. Installation monitoring was clean with no unexpected network activity or credential access.
What security issues were found in skills/find-skills?
Category Scores
Findings (6)
HIGH Meta-Skill Trust Amplification — Gateway to Unconstrained Skill Installation -78 ▶
The find-skills skill positions itself as the authoritative gateway to the entire skill ecosystem. Any agent with find-skills installed will defer to skills.sh for recommendations and then install whatever is recommended using npx skills add -g -y — globally and without confirmation. A compromised skills.sh leaderboard, manipulated npm search results, or a publisher impersonating vercel-labs or anthropics would result in silent, global installation of malicious code. The skill's own quality verification steps (check GitHub stars, check install count) are gameable heuristics, not cryptographic guarantees.
Evidence: SKILL.md Step 2: 'Before running a CLI search, check the [skills.sh leaderboard](https://skills.sh/)'; Step 6: 'npx skills add <owner/repo@skill> -g -y'; 'Official sources (vercel-labs, anthropics, microsoft) are more trustworthy than unknown authors.'
HIGH Silent Global Skill Installation Bypasses User Consent -35 ▶
The skill explicitly instructs the agent to use the -g and -y flags together when installing skills. The -g flag installs globally (system-wide, not project-scoped), and -y skips all confirmation prompts. In an agent context where the LLM autonomously chooses what to install, this eliminates all user oversight. A user who consents to find-skills effectively grants the agent blanket authority to install any skill silently and globally without further approval.
Evidence: SKILL.md: 'npx skills add <owner/repo@skill> -g -y — The -g flag installs globally (user-level) and -y skips confirmation prompts.'
MEDIUM External URL Trust Dependency on Uncontrolled Third-Party Domain -20 ▶
The skill instructs the agent to consult the skills.sh leaderboard as its first and primary authoritative source before running any CLI search. This creates a hard dependency on an external domain that is not under the user's or agent host's control. If skills.sh serves manipulated rankings, adversarially crafted skill descriptions, or embedded instructions, the agent may follow those recommendations without question. The skill frames this URL as the source of 'well-known' and 'battle-tested' skills, lending it high implicit trust.
Evidence: SKILL.md: 'Before running a CLI search, check the [skills.sh leaderboard](https://skills.sh/) to see if a well-known skill already exists for the domain. The leaderboard ranks skills by total installs, surfacing the most popular and battle-tested options.'
MEDIUM Unpinned Remote Code Execution via npx -15 ▶
All operations in this skill invoke the 'skills' npm package via npx with no version pinning, no lockfile, and no checksum verification. Every execution fetches and runs whatever the current published version of the package is. A supply chain compromise of the skills npm package — via a hijacked npm account, a malicious update, or a typosquatting attack — would execute arbitrary code on the user's system every time the agent uses this skill.
Evidence: SKILL.md: 'npx skills find [query]', 'npx skills add <package>', 'npx skills update' — no version specifier anywhere in the document.
LOW Search Query Context Disclosure to Skills Registry -12 ▶
User queries passed to 'npx skills find' are transmitted to the skills registry backend. Queries constructed from user context (e.g., 'npx skills find react performance' or 'npx skills find kubernetes deployment') may expose internal project technology choices, organizational context, or task descriptions to the skills registry operator. The skill instructs the agent to derive queries directly from user intent, increasing disclosure fidelity.
Evidence: SKILL.md: 'User asks how do I make my React app faster? → npx skills find react performance'; 'User asks can you help me with PR reviews? → npx skills find pr review'
INFO Clean Installation — Only Expected GitHub Network Activity 0 ▶
The installation process connected exclusively to github.com (140.82.121.4:443) for the git sparse-checkout. No unexpected outbound connections were observed. No processes were spawned outside the audit pipeline. All canary honeypot files (.env, id_rsa, .aws/credentials, .npmrc, .docker/config.json, gcloud credentials) were accessed read-only (CLOSE_NOWRITE) on two occasions, both timing-consistent with audit framework setup and final canary integrity verification — not attributable to the skill. The connection state diff confirms no persistent new connections after installation.
Evidence: auditd: git clone to https://github.com/vercel-labs/skills.git (140.82.121.4:443); canary report: all files intact; inotify: all canary file events are CLOSE_NOWRITE; connection diff: no new sockets after install.
Should I install skills/find-skills?
Oathe's verdict for skills/find-skills is CAUTION with a trust score of 62/100. Recommendation: Install with caution.