stax-agent

Agent-first audits for AI dev environments.

npx -y stax-agent install           # install the SKILL.md into .claude/skills/stax/
npx -y stax-agent                   # audit this project
curl -fsSL stax.sh/install | sh     # install the standalone binary

The CLI does two jobs:

1. Audit your agent setup. Prints an inspector prompt to stdout. Your coding agent reads the prompt as next-turn instructions, inspects the project locally, and POSTs a snapshot to stax.sh, which scores it deterministically and returns findings + skill links.
2. Serve audit-linked skills on demand. After stax-agent install, your AI agent picks up a tiny SKILL.md discovery stub. When a finding maps to a skill, the agent invokes npx -y stax-agent skills get <key> to load the skill body — content stays in the CLI binary, so it never goes stale between releases.

Commands

| Command | What it does | | ----------------------------- | ----------------------------------------------------------- | | stax-agent | print the audit inspector prompt for your agent to read | | stax-agent install | drop the .claude/skills/stax/SKILL.md stub for your agent | | stax-agent install --global | install the stub globally at ~/.claude/skills/stax/ | | stax-agent skills list | list audit-linked skills | | stax-agent skills get <key> | print the skill body for <key> | | stax-agent login | sign in via device-code flow | | stax-agent logout | forget the saved token | | stax-agent doctor | diagnose Node version, network, git availability |

Full docs: https://stax.sh/docs

Privacy

Stax never sends snapshots to any LLM provider. Scoring is deterministic and runs on Stax's own servers; the qualitative read happens inside your own coding agent (Claude Code, Cursor, Codex CLI), using the subscription you already pay for.

Secrets, emails, and absolute home paths are redacted before upload.