Escalation format for AI agents: decisions in seconds, findable later
A fixed format for AI coding agents to stop and ask a human: stable id, three options, a recommendation, what happens if nobody answers, and the decisions log entry it produces.
How to use it
- Put the format in CLAUDE.md so every session knows it.
- Answer with the id and a letter: "014-3 B". Anything longer belongs in the log entry.
- The lead agent writes the log entry, citing the id. A decision that only lives in a chat is gone next session.
DECISION <ticket>-<n>: the question in one line
A) option B) option C) option
RECOMMENDATION: <letter>. WHY: one line. IF UNDECIDED: what happens
DECISIONS REQUIRED: <id>, <id>
ACTIONS FOR YOU: <what to open or check, with the URL>## <ticket>-<n>: <the question in one line>
Date: YYYY-MM-DD
Asked by: <agent role> while working on ticket <id>
Options: A) ... B) ... C) ...
Decided: <letter>, by <person>
Why: <one line, from the recommendation or the answer>
Consequences: <what this now rules in or out; tickets that must cite it>Prompt to use it
When you reach a decision that changes what the product does, who can see what, or what we store, stop and escalate.
Do not guess and do not write paragraphs. Use exactly this format:
DECISION <ticket>-<n>: the question in one line
A) option B) option C) option
RECOMMENDATION: <letter>. WHY: one line. IF UNDECIDED: what happens
Group every open decision at the end of your reply under DECISIONS REQUIRED.
Implementation choices that change nothing a user or a role would notice are yours: take them and mention them in one line.Why it looks like this, and where each part came from: Human in the loop for coding agents: the two gates.
Changelog
- v1.0Nov 5, 2026First public version. Over seven weeks this format carried more than 170 decisions, most answered in under a minute, every one findable later by id.
More from the kit
CLAUDE.md template
The seven sections every agent session inherits, in the order an agent reads them, with nine neutral hard rules to start from.
v1.0 / CLAUDE.mdTemplate / BuildTicket template for agents
Frontmatter a script can validate, plus the three fields that stop agents inventing requirements. Blank and filled versions.
v1.0 / ticket.mdChecklist / BuildRelease gate checklist
The merge gate and the release gate, as checkboxes. Written for the day a skipped check reports green.
v1.0 / release-gate.mdWorking on something like this?
This is the method behind how I build: AI native development with human gates. Discuss a product idea → Or email me.