Issue triage

Clear an issue backlog in parallel: a script hands each open issue to its own agent, each one investigates read-only and drafts a triage note, and one agent merges them into a single list you can act on.

Pull the backlog

With the GitHub CLI, for example:

gh issue list --label needs-triage --limit 10 --json number,title --jq '.[].number' > issues.txt

Fan out, one agent per issue

#!/usr/bin/env bash
set -uo pipefail
repo=$(pwd)
mkdir -p reports/triage
while read -r n; do
  beswarm agent start --name "issue-$n" --agent claude --group triage --no-focus \
    --prompt "In $repo: read GitHub issue #$n with 'gh issue view $n --comments'. Investigate the code read-only — don't edit files or comment on the issue. Write to $repo/reports/triage/$n.md: a one-line summary, likely cause with file:line, severity (low/medium/high), a size estimate (S/M/L), and whether it's a duplicate of another issue."
done < issues.txt
 
while read -r n; do
  beswarm agent wait "issue-$n" --until done,blocked --timeout 1800
done < issues.txt

Ten agents at once is a lot for one machine and one API quota. Start with a handful and batch the rest.

Merge into one list

beswarm agent start --name triage-lead --agent codex --group triage \
  --prompt "Merge the notes in reports/triage/ into reports/triage/summary.md: a table sorted by severity then size, with duplicates grouped. Don't post anything to GitHub." \
  --wait --timeout 1200

Act on it

Read the summary in the Artifacts tab, then label, assign or close issues yourself. Agents draft; you decide what gets posted.

beswarm agent close --group triage