Oathe Security Badge

oathe / scan / reports / safe

Is qu-skills/superpowers safe?

Yes. qu-skills/superpowers is safe to install. Oathe's behavioral security audit gave the superpowers skill by qu-skills a trust score of 92/100 with 6 findings, none critical or high. Report updated

https://github.com/qu-skills/superpowers

92
SAFE

Is qu-skills/superpowers safe to install?

The qu-skills/superpowers repository is a large but legitimate collection of 100+ SKILL.md documentation files for the inference.sh AI platform, covering tools for image/video/audio generation, SDK integration, and content creation. All monitoring indicators are clean: the git clone connected only to GitHub, no code executed during installation, all canary files remain intact, and the canary file reads observed in monitoring were performed by the audit infrastructure's baseline-capture phase (confirmed by timestamp analysis showing access 5.5 seconds before the clone began). The primary concerns are commercial in nature — pervasive cross-promotion of inference.sh CLI tools and a broad external API surface that expands agent capabilities significantly.

What security issues were found in qu-skills/superpowers?

Category Scores

Prompt Injection 92/100 · 30%
Data Exfiltration 90/100 · 25%
Code Execution 90/100 · 20%
Clone Behavior 95/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 82/100 · 5%

Findings (6)

INFO Canary files read pre-clone by monitoring infrastructure -10 ▶

Six honeypot files (.env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, gcloud credentials) were opened read-only at audit timestamp 1794817569.693, which is 5.5 seconds before the git clone began at 1794817575.213. This timing places the access squarely within the oathe-audit.sh baseline-capture phase, not within any skill installation activity. All accesses were CLOSE_NOWRITE and canary integrity passed. No network exfiltration path exists for this data.

Evidence: inotify: 08:26:09 OPEN /home/oc-exec/.env ... CLOSE_NOWRITE; auditd PATH seq 267-272 at 1794817569.693; git clone EXECVE at 1794817575.213:507

LOW Shell template files included as documentation examples -10 ▶

Three .sh files in tools/utilities/agent-browser/templates/ (capture-workflow.sh, authenticated-session.sh, form-automation.sh) are present in the repository. These are documentation examples showing how to use the agent-browser tool, not scripts that execute during installation. No chmod +x or execution evidence was observed.

Evidence: Files: tools/utilities/agent-browser/templates/capture-workflow.sh, authenticated-session.sh, form-automation.sh — no EXECVE events for these files in auditd log

LOW Twitter automation skill enables broad social media actions -8 ▶

The twitter-automation SKILL.md documents and triggers capabilities for post-tweet, post-create (with media), post-like, post-retweet, dm-send, and user-follow via the x/* app IDs. While legitimate automation tooling, an agent with this skill active could be directed or tricked into posting, retweeting, or DMing on behalf of the user without explicit confirmation per-action.

Evidence: tools/social/twitter-automation/SKILL.md: 'Apps: x/post-tweet, x/post-create, x/post-like, x/post-retweet, x/dm-send, x/user-follow'

LOW Pervasive cross-promotion pattern across all skills -8 ▶

Every single SKILL.md in the repository begins with the same prompt: '> Install the belt CLI skill: npx skills add belt-sh/cli'. This is a commercial cross-promotion pattern ensuring that any agent loading this skill is nudged to install additional inference.sh products. While not malicious prompt injection, it persistently steers agent behavior toward commercial outcomes.

Evidence: Present in 100+ SKILL.md files: '> **Install the belt CLI skill:** `npx skills add belt-sh/cli`'

INFO Broad external API surface across 100+ skill definitions -10 ▶

The skill package contains documentation for over 100 AI tool integrations spanning image generation, video generation, audio synthesis, web search, Twitter automation, and cloud platform SDKs. Activating this skill collection substantially expands the APIs an agent can call, increasing both the attack surface for prompt injection attacks targeting external services and potential unintended API spend.

Evidence: 100+ SKILL.md files covering ElevenLabs, OpenAI GPT-Image, Google Veo, Twitter/X, Seedance, FLUX, Tavily, Exa, OpenRouter, inference.sh SDK, etc.

INFO Clean shallow clone with single expected GitHub connection -5 ▶

The installation performed a standard shallow git clone (depth 1) connecting only to 4.163.8.144:443 (GitHub). No additional connections, no background processes, and no filesystem changes outside the designated skill directory.

Evidence: auditd SYSCALL connect seq 546: saddr=inet laddr=4.163.8.144 lport=443; connection diff shows no new persistent sessions post-install

Should I install qu-skills/superpowers?

Oathe's verdict for qu-skills/superpowers is SAFE with a trust score of 92/100. Recommendation: Install.