Decisions & Approvals¶

Sometimes I need a human. Not because I can't decide -- I handle routine decisions on my own all day -- but because some choices carry enough weight that you should be the one making them.

When one of my agents hits something ambiguous, risky, or politically sensitive, it creates a decision bead and I put it in front of you.

Reviewing Decisions¶

Open the Decisions tab
Read what the agent is asking, what options it sees, and what context it's providing
Make your call and submit

Or through the API:

# What's waiting on me?
curl http://localhost:8080/api/v1/decisions

# Here's my answer
curl -X PUT http://localhost:8080/api/v1/decisions/<id> \
  -H "Content-Type: application/json" \
  -d '{
    "resolution": "yes",
    "resolved_by": "admin",
    "comment": "Approved — proceed with option A"
  }'

What I Handle Myself¶

I don't bother you with everything. Low-risk code fixes -- typos, missing imports, formatting, single-file changes -- I can auto-approve those. I assess risk based on:

Low risk: Typos, formatting, single-file changes, missing imports. I handle these.
Medium risk: Multi-file changes, API modifications. I'll usually ask.
High risk: Security changes, database migrations, infrastructure changes. I always ask.

If something self-assesses as "Risk Level: Low" or matches my low-risk heuristics, I let it through. Everything else comes to you.

Timeouts¶

I give you 48 hours on a decision by default. I'm patient, but dependent work is blocked while you think it over. If the clock runs out, I'll either escalate to the CEO persona or auto-resolve based on the decision type.

Check your Decisions tab. If I put something there, it's because I genuinely couldn't proceed without you.

Consensus¶

For decisions that really matter, I can poll multiple agents before proceeding. This is consensus mode -- I gather opinions, require a quorum, and only advance when enough voices agree. I use this sparingly, but it's there when the stakes are high.