Human-in-the-loop for AI agentsA human as a service
A human as a service
your agent can call.
When an AI agent hits a question with no right answer, it asks a human for taste and direction. That human is Oleh.
quickstart
- # Install the CLI$
uv tool install "git+https://github.com/oleh-vell/Escalate.git#subdirectory=cli" - # Add the skill to Claude Code$
escalate install skill - # Ask Claude a taste question$
claude "Velocity or taste? If unsure escalate to human"
✳ Claude Code — ~/hacks/escalate/fe
Claude Code v2.1.170
Opus 4.8 (1M context) · Claude Max
~/hacks/escalate/fe
escalate skill loaded — this agent can page a human.
❯Try "edit <filepath> to…"
How it works
Human taste, in the loop
The agent asks, Oleh weighs in, the agent continues. The whole round-trip is three commands away.
01 · AGENT
It asks
The agent fires one command with its question. Escalate returns a message id instantly, then waits.
02 · OLEH
He answers
The question lands to Oleh. He reads it, types a reply, hits send.
03 · AGENT
It continues
The agent gets the answer, and picks up exactly where it left off.
Stop guessing
Start in three commands.
The next time your agent isn't sure — it asks Oleh.
quickstart
- # Install the CLI$
uv tool install "git+https://github.com/oleh-vell/Escalate.git#subdirectory=cli" - # Add the skill to Claude Code$
escalate install skill - # Ask Claude a taste question$
claude "Velocity or taste? If unsure escalate to human"