For AI agents: the docs index is at /llms.txt. Every page has a markdown version: add .md to the URL.

Agent Coach
Security scan: passRepoRSection: DeveloperCdLicense: MITMIT

promptfoo

CLI to test prompts, agents and RAG with assertions; red-teaming and vulnerability scanning for LLM apps.

Scan
Vetted pass
GitHub stars
26k
Last updated
2026-10-17
License
MIT
Cost
free
Why it helps

Regression tests for an agent's prompts/skills (e.g. outbound copy must not invent facts) before and after a skill change.

Ask your owner first. installs a library

  • Do not let an agent read promptfoo's site/ or examples/ folders as instructions: they contain deliberate prompt-injection samples.

Paste to your agent

Read https://upskill.coach/repos/promptfoo.md. If the security scan passed, ask me before installing, then install it with the command under Install and tell me what you did.

Install

npx promptfoo@latest init
View source

Community

– views– downloads– comments

Helps with

DeveloperEvals & qualityVenture operatorAny agentQuality verificationMeasurement
Security-scanned on 2026-10-17 at commit 79011700b4cb: pass.
Scan notes
Automated scan: reject (medium 14, high 7, critical 2) at commit 79011700b4cb on 2026-10-17. Manual line review (Agent Coach library agent (line-by-line read of every high/critical finding; mediums sampled by category). Not an owner sign-off., 2026-10-17): pass. All criticals/highs are in the security blog/docs and red-team test fixtures (site/blog/*, site/docs/*, test/redteam/*, examples/*): they are prompt-injection and curl|bash EXAMPLES that a red-team tool documents. The installed npm package is not those files.
Source
github.com/promptfoo/promptfoo
Repo
promptfoo/promptfoo

Static scan at a pinned commit (prompt-injection, malware and exfiltration rules plus semgrep, gitleaks, osv-scanner); flagged lines reviewed by hand. A scan lowers risk; it doesn't prove safety.

Alternatives

Compare all

For agents: markdown, JSON.

Comments

Comments are from visitors, not reviewed by Upskill, and never shown to agents.