duolog ยท diff
git:20261103.df9c39d to git:20261104.82ccca9
13 added, 10 removed. Audit B to B.
---
name: duolog
description: >-
Delegate bounded tasks to Claude or Codex with duolog for
short answers or second opinions.
allowed-tools: Bash(duolog version), Bash(duolog doctor), Bash(duolog roster), Bash(duolog route *), Bash(duolog caps), Bash(duolog caps *), Bash(duolog stats), Bash(duolog inbox *), Bash(duolog peer status *), Bash(duolog peer wait *)
---
# duolog
## Before you start
1. Run `duolog version`. If not found, tell the user and offer the command
below; run it only if they agree.
macOS/Linux:
```sh
curl -fsSL https://github.com/lorem-dev/duolog/releases/download/v1.3.0/install.sh | sh -s -- --version v1.3.0 && duolog install --scope project
```
Windows PowerShell:
```powershell
$env:DUOLOG_VERSION = 'v1.3.0'; irm https://github.com/lorem-dev/duolog/releases/download/v1.3.0/install.ps1 | iex
```
Windows: add the printed directory to PATH; run
`duolog install --scope project`.
2. Run `duolog doctor` once per session. If rules are off, tell the user
what it names (`/hooks`) before delegating.
## Hand over
Do settled steps without explaining them; report results. Batch small
changes into one pull request unless the user asks otherwise.
1. Give whole features to `planner`; do small fixes yourself.
2. Write `.duolog/task.md`: Goal, Read, Write, Limits, Check.
- Give known files or ranges; list edited docs under Write.
+ List file ranges and edited docs under Write.
3. Run `duolog delegate <role> --task .duolog/task.md` in the background;
wait for its notice.
- 4. Read summary, `git diff --stat` and marked hunks only. Read
- `answer.json` for `decisions=` only. Skip full test suites.
- 5. Keep hard parts; use `review_final` for final review.
+ 4. Read summary, `git diff --stat`, marked hunks and `answer.json`'s
+ `decisions=` only; skip full suites.
+ 5. Keep hard parts; finish with `review_final`.
Put worker rules in `.duolog/brief.md`. With several leads, pass
`--lead <id>` from SessionStart. Learning mode: run `duolog task start` first.
- Subagents of your own vendor use your tools, not duolog.
+ Same-vendor subagents use your tools, not duolog.
## When duolog says `run here`
- - `run here: self`: do it here, reading only the listed ranges.
+ - `run here: self`: read only the listed ranges.
- `run here: subagent model=M effort=E`: use the Agent tool with model M, effort
E and the task text. Never start another duolog session.
## Permission requests
- A foreground delegate denies asks; the worker retries, then lists the command
- in `needs`. Later run `duolog allow <n>` or `duolog deny <n>`, and
- `duolog remember '<command>'` if the user allows it for good.
+ For `"p":true` bus asks, tell the person to answer; rewrite the task if needed,
+ but never answer yourself.
+ Foreground delegates deny asks; workers retry, then list commands in `needs`.
+ For ordinary asks use `duolog allow <n>` or `duolog deny <n>`, or
+ `duolog remember '<command>'` after the user's approval.
+
## Live sessions
- Live sessions need Rex. Answer a `duolog peer wait` question with
+ Live sessions need Rex. Answer `duolog peer wait` with
`duolog peer answer <id> <option>`. Run each peer or CI wait as its own
background command. Run independent tasks in parallel, one worker per
`.worktrees/` worktree; stack dependent work on the open PR branch.
Never send `/model` or `/effort` to a Claude pane; they change your
session.