PrismaLens

Investigate incidents with Claude Code

Codex, OpenCode and Gemini CLI work the same way.

curl -fsSL https://prismalens.io/install.sh | sh

Then run pl up

Also

Agent setup → 
$ PRISMALENS_HARNESS=claude-code pl up

Can Claude Code investigate incidents?

Yes. PrismaLens drives Claude Code on your machine over the Agent Client Protocol, the way an editor does. Claude Code reads the service's code and writes a report that ranks its hypotheses with evidence, using its own sign-in. Codex, OpenCode, Gemini CLI and deepagents work the same way.

From alert to findings

  1. Install PrismaLens

    One command, then run pl up.

  2. Sign in your agent

    It uses the agent's own sign-in, as in your terminal.

  3. Pin Claude Code

    Start with PRISMALENS_HARNESS=claude-code, or choose it in settings.

  4. Hand it an incident

    Send an alert, or write the incident yourself.

Questions, answered

Does PrismaLens see my Claude Code login?

No. It never reads or stores the agent's own login. If you enter an API key instead, PrismaLens stores it encrypted and never shows it again.

What may Claude Code do during a run?

You pick one of four permission levels. The default asks before anything beyond reading and searching. A run started by an alert runs at Auto and refuses anything the agent asks. PrismaLens adds no sandbox.

How is Claude Code picked over other agents?

PRISMALENS_HARNESS=claude-code pins it. Otherwise the agent chosen in Settings → Agent runs, and failing that, the first agent that can run.