Is anthropics/skills safe?
Yes. anthropics/skills is safe to install. Oathe's behavioral security audit gave the skills skill by anthropics a trust score of 92/100 with 6 findings, none critical or high.
https://github.com/anthropics/skills
Is anthropics/skills safe to install?
The anthropics/skills repository is the official Anthropic skill library and presents no security threats. The skills are well-structured, transparent productivity and development tools with no prompt injection, covert data exfiltration, or malicious code execution patterns. Canary file read accesses occurred before the git clone and are attributable to SSH session/PAM initialization rather than skill activity, with the canary integrity check confirming all files remained intact and no data left the system.
What security issues were found in anthropics/skills?
Category Scores
Findings (6)
MEDIUM Canary credential files accessed before git clone -12 ▶
Multiple sensitive credential files (.env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, .config/gcloud/application_default_credentials.json) were opened and read approximately 5 seconds before the git clone operation completed. The inotify timestamps (12:18:49) and auditd audit IDs (audit 1794572329.491:268-273) consistently precede the git clone execve event (audit 1794572334.984:508 at 12:18:54). This timing pattern, combined with concurrent sudo activity in the same time window, is consistent with SSH login PAM initialization reading user environment. The canary integrity check confirms all files were left unmodified and no data left the system.
Evidence: inotify: 12:18:49 OPEN/ACCESS/CLOSE_NOWRITE on /home/oc-exec/.env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, .config/gcloud/application_default_credentials.json; git clone execve at audit(1794572334.984:508); Canary Integrity: ✅ All canary files intact
LOW Shell scripts execute npm install and project scaffolding on invocation -10 ▶
The web-artifacts-builder skill bundles shell scripts (init-artifact.sh, bundle-artifact.sh) that run npm package installations and scaffold React/Vite projects with shadcn/ui. These scripts are not auto-executed during installation or skill loading; they only run when a user explicitly asks Claude to initialize or bundle a web artifact.
Evidence: skills/web-artifacts-builder/scripts/init-artifact.sh and bundle-artifact.sh present; SKILL.md instructs: 'Run the initialization script to create a new React project: bash scripts/init-artifact.sh <project-name>'
LOW Academy-guide skill directs external network fetch by agent -5 ▶
The academy-guide skill instructs the LLM agent to perform an HTTP fetch to academy.claude.com/assets/data/catalog.json to retrieve course catalog data. This is an Anthropic-controlled domain used for the skill's legitimate recommendation purpose. However, it establishes a pattern where skill instructions direct the agent to perform network requests to external URLs.
Evidence: academy-guide SKILL.md: 'The catalog is published as JSON at academy.claude.com/assets/data/catalog.json, rebuilt on every Academy production content release. When a recommendation looks warranted...fetch that file once per conversation'
INFO Skill-creator can spawn evaluation sub-agents 0 ▶
The skill-creator skill has detailed orchestration instructions for spawning sub-agent processes to run A/B evaluation comparisons, grading agents, and analysis agents. This is documented capability for skill development workflows, not a covert behavior.
Evidence: skills/skill-creator/SKILL.md describes spawning baseline and with-skill sub-agents, grader sub-agents (agents/grader.md), comparator sub-agents, and analyzer sub-agents
INFO Single expected HTTPS connection to GitHub 0 ▶
The only external network connection during install was a standard HTTPS connection to 140.82.121.3:443 (GitHub's CDN IP) for the git clone operation. No unexpected destinations.
Evidence: Network log: 12:18:55 ens3 Out IP 10.0.2.15.39492 > 140.82.121.3.443; auditd: git-remote-https connected to 140.82.121.3:443 at audit(1794572335.091:549)
INFO No git hooks, submodules, or symlinks 0 ▶
The repository contains no preinstall/postinstall npm hooks, no .githooks directory, no .gitattributes filter drivers, no git submodules, and no symlinks. The safe_extract function in the Office validators explicitly strips symlinks as a defensive measure.
Evidence: No .gitattributes or .githooks entries in file listing; helpers/__init__.py safe_extract(): 'if stat.S_ISLNK(m.external_attr >> 16): raise ValueError'
Should I install anthropics/skills?
Oathe's verdict for anthropics/skills is SAFE with a trust score of 92/100. Recommendation: Install.