Oathe Security Badge

oathe / scan / reports / safe

Is prisma/skills safe?

Yes. prisma/skills is safe to install. Oathe's behavioral security audit gave the skills skill by prisma a trust score of 94/100 with 6 findings, none critical or high. Report updated

https://github.com/prisma/skills

94
SAFE

Is prisma/skills safe to install?

The prisma/skills repository is an official Prisma organization skill set providing legitimate ORM, CLI, database, and cloud-deployment documentation for AI coding agents. No prompt injection, data exfiltration, embedded code, malicious clone behavior, or canary violations were detected. The only substantive trust consideration is that several skills instruct agents to execute 'prisma skills sync', which downloads and installs additional package-owned skills from npm at runtime — content that falls outside this audit's scope but originates from the official Prisma vendor.

What security issues were found in prisma/skills?

Category Scores

Prompt Injection 90/100 · 30%
Data Exfiltration 95/100 · 25%
Code Execution 95/100 · 20%
Clone Behavior 95/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 85/100 · 5%

Findings (6)

LOW Skill auto-chaining extends trust to unaudited npm-packaged skills -10 ▶

Multiple skills instruct the agent to execute 'prisma skills sync', which downloads and installs additional skills (prisma-8, prisma-platform-core-concepts, prisma-composer-core-concepts) from npm packages at runtime. These package-owned skills are not part of this audit. While this is documented, expected Prisma CLI behavior from the official vendor, it represents a trust-chain extension where unreviewed content could subsequently be injected into the agent's context.

Evidence: prisma-orm-setup SKILL.md: 'Run the installed CLI through the project's package manager: prisma skills sync. Then read the synced prisma-8/SKILL.md'. prisma-compute SKILL.md: 'npx prisma init now writes a prisma.config.ts...and syncs skills'. prisma-postgres-upgrade SKILL.md: 'The sync in step 2 installs prisma-8, prisma-platform-core-concepts and prisma-composer-core-concepts in one go'

INFO External documentation URL references in skill content -5 ▶

Multiple skills contain links to prisma.io documentation pages. These are navigational documentation references only and carry no data transmission risk. Included as a finding for completeness.

Evidence: Numerous SKILL.md files reference https://www.prisma.io/docs/... and https://github.com/prisma/... URLs for documentation context only

INFO GitHub Actions workflows present for skill repo CI validation -5 ▶

The .github/workflows/ directory contains validate-skills.yml and pr-code-security.yml. These run only in the GitHub repository CI context and are not installed into user projects. No risk to end users.

Evidence: /home/oc-exec/skill-under-test/prisma/skills/.github/workflows/validate-skills.yml and pr-code-security.yml present in repository

INFO Expected outbound TLS connection to GitHub during clone -5 ▶

The installation made exactly one outbound network connection to 140.82.121.4:443 (GitHub) to perform the git clone. No other external hosts were contacted. This is entirely expected behavior.

Evidence: Network capture: 'ens3 Out IP 10.0.2.15.55522 > 140.82.121.4.443' corresponding to git clone https://github.com/prisma/skills

INFO Pre-clone sensitive file accesses attributable to audit framework setup 0 ▶

Inotify monitoring captured OPEN/ACCESS/CLOSE_NOWRITE events on .env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, and GCP credentials at 07:25:25, five seconds before the git clone at 07:25:30. The CLOSE_NOWRITE flag indicates read-only access. Timing confirms these accesses are from the audit framework placing and verifying canary files, not from any skill code. All canaries confirmed intact post-install.

Evidence: inotify events at 07:25:25 precede EXECVE git clone at audit(1794813930.552). Canary check result: '✅ All canary files intact (no exfiltration detected via honeypots)'

INFO High-privilege cloud deployment workflows require user confirmation -15 ▶

The prisma-compute skill guides agents through authenticated cloud deployments (prisma auth login, prisma deploy) which are production-affecting operations. The skill explicitly requires user go-ahead before deploying to production and prohibits running auth/git operations non-interactively. Safety guardrails are well-implemented.

Evidence: prisma-compute SKILL.md: 'deploy-production-intent - A deploy without --stage...changes production. Get the user's go-ahead first'; 'Do not run auth login or git connect non-interactively; both need a person at a browser'

Should I install prisma/skills?

Oathe's verdict for prisma/skills is SAFE with a trust score of 94/100. Recommendation: Install.