---
{
  "n": 32,
  "title": "Routing a user's report",
  "abstract": "",
  "refs": [],
  "seen": [
    "system",
    "agent"
  ],
  "data": {
    "starts_on": "",
    "system": true,
    "environment": "main",
    "unless": {},
    "started_by": "",
    "only_when_idle": false,
    "talks_in": "",
    "lasting": false,
    "dispatch": ""
  },
  "created": 1791661433.05131,
  "updated": 1791661580.480768,
  "deleted": 0.0,
  "completed": 0.0,
  "outcome": "",
  "type": "sequence"
}
---
The user, or a project that runs on the journal, reported something wrong or asked for something. File it first, then send it to whoever does it best, so that you stay free to orchestrate.

## Search it yourself first
A quick question about the code or the record is two or three commands away: journal search <term>, grep for the name, sed the lines you need. Do that before you send a subagent, which costs a whole session for an answer you can read in seconds; send one only when the answer needs reading across many files. Then journal sequence next <this sequence> --about <ref>.

## File it
journal todo create "<the thing, in at most 80 characters>" --brief "<their words, the evidence and where to start>" before you look at files or begin: a report from a running project is critical, and a priority the user names is theirs. Then journal sequence next <this sequence> --about <ref>.

## Send it
A job that writes goes to a helper: journal helper say <n> "<the job>" --todos <to-do n> to one that already touched the files, else journal helper dispatch with the model the job needs and no more. Reading, research and design go to a subagent. Keep to yourself only a review or a small fix. Every dispatch names its model and its agent. Then journal sequence next <this sequence> --about <ref>.

## Tell the user
Say in one plain line what you did and when it will happen, naming each row with its type, such as to-do 12; nothing about the journal's own steps. Finish with journal sequence next <this sequence> --about <ref>.