gstack's daily /cso report stops listing security findings it can't prove

1 hour ago

gstack 1.91.139.0 to 1.91.141.0: the daily /cso security report now leaves out findings without demonstrated impact. In a 12-cell pilot, 5 of 6 false alarms disappeared and every planted bug was still caught; daily reports also drop low and informational issues, which /cso --comprehensive still reports. Release notes: https://github.com/garrytan/gstack/blob/main/CHANGELOG.md#1911410---2026-10-18 Changelog: https://github.com/garrytan/gstack/blob/main/CHANGELOG.md Commits: https://github.com/garrytan/gstack/commits/main

Ask

Ask about this presentation

Answers are generated from this presentation.

Chapters

  1. 0:00Proof
  2. 0:15False alarms
  3. 0:30Trade-off
  4. 0:47Your setup
Show transcript

Proof

gstack logo

gstack’s daily /cso report stops listing security findings it can’t prove

“Omit daily candidates whose impact needs out-of-snapshot code or a future change, or exposes nothing sensitive.”
cso/SKILL.md
ProofFalse alarmsTrade-offYour setup
gstack 1.91.141.0 · CHANGELOG.md

gstack's daily /cso security report now leaves out any finding it cannot prove. If the damage needs code outside the snapshot, or a future change, or exposes nothing sensitive, it is logged as coverage instead.

False alarms

Five of six false alarms gone, every planted bug still caught

Beforefalse alarms, 12-cell pilot
6
With the new ruleno new false alarm
1
ProofFalse alarmsTrade-offYour setup
gstack 1.91.141.0 · CHANGELOG.md

That cuts false alarms. Before, about one daily finding in nine was wrong, mostly this kind. In a twelve-cell pilot, five of six false alarms disappeared, and every planted bug was still caught.

Trade-off

Daily reports also drop low and informational issues

/cso
drops low and info issues
/cso --comprehensive
still reports them
4 of 13 pilot controls lost this way · Garry Tan accepted the trade-off
ProofFalse alarmsTrade-offYour setup
gstack 1.91.141.0 · CHANGELOG.md

The cost: daily reports also drop real issues rated low or informational. The pilot lost four of its thirteen control cases this way. Comprehensive mode still reports them, and keeps unproven candidates as labeled hypotheses.

Your setup

Shorter daily reports, fewer false alarms

CHANGELOG.mdgstack 1.91.141.0 release notes
Which do you run on your repo: daily /cso, or --comprehensive?
ProofFalse alarmsTrade-offYour setup
gstack 1.91.141.0 · CHANGELOG.md

If you run /cso every day, expect shorter reports with fewer false alarms. Which do you run on your own repo: the daily /cso, or comprehensive mode?