AevrinAI Agent Prompts
Precise prompts for installing the CLI and safely merging the hook configuration.
These prompts tell an agent to use supported commands, preserve existing configuration, and stop instead of guessing. Review every requested change before allowing it to proceed.
Install the CLI
Set up the Aevrin CLI on this machine and verify the installation end to end.
1. Detect the OS and shell. Do not use sudo or modify the system Python.
2. Install pipx with the platform-appropriate command, then run pipx install aevrin.
3. If Aevrin already exists, run pipx upgrade aevrin instead.
4. Run aevrin --version and aevrin login.
5. Pause while I approve browser device login. Never ask for a password, API key, or token.
6. Show the exact commands and results. Stop on the first failure and preserve its non-secret error.Configure the project hook
Configure Aevrin's Claude Code PreToolUse hook without overwriting existing settings.
1. Run aevrin --version, then aevrin hook setup.
2. Pause while I approve browser device login. Never print the saved credential.
3. Merge the exact Bash and Write matcher entries printed by Aevrin into .claude/settings.json.
4. Preserve every existing key and hook. Create the file only if it does not exist.
5. Validate the final file as JSON and show the non-secret configuration and path changed.
6. Stop if setup fails, the JSON cannot be merged safely, or a matcher conflicts.Review a JSON report
Review this Aevrin JSON report without running its target. Summarize incomplete or skipped
coverage first. Group open findings by severity and OWASP MCP category. For each critical or high
finding, cite file/line evidence, explain impact, propose the smallest fix, and give a repeat-scan
verification command. Do not mark anything false-positive without specific evidence.