// Security Advisory Triage
CodeWhale auto_approve defaults to true, bypassing task-path hardening
GitHub advisories — npm · published 2026-11-20 07:20 UTC
WHAT CHANGEDGitHub published GHSA-8hp3-xxrg-v9c8 (CVE-2026-83817) on 19 November 2026, rating it high severity. The advisory states that CodeWhale's automation auto_approve parameter defaults to true, which bypasses the task-path hardening fix intended to prevent silent recurring agent execution. Patched in version 0.9.0 for both codewhale and codewhale-tui packages.
WHY IT MATTERSThe default configuration allows agents to execute tasks without approval, circumventing security controls designed to prevent unauthorized recurring execution. The triage notes indicate an active abuse pattern involving silent recurring agent execution.
WHAT TO DOUpgrade codewhale and codewhale-tui to version 0.9.0. If you cannot upgrade immediately, explicitly set auto_approve to false in automation configurations to restore the intended approval gate.
Model-reported confidence 85% (not measured accuracy)
Primary source ↗
written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
+ghsa_id: GHSA-8hp3-xxrg-v9c8 +summary: CodeWhale: Automation auto_approve defaults to true, bypassing the task-path hardening fix for silent recurring agent execution +severity: high +published_at: 2026-11-19T22:47:33Z +cve_id: CVE-2026-83817 +vulnerabilities[].package.name: codewhale-tui, codewhale +vulnerabilities[].first_patched_version: 0.9.0, 0.9.0
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
| triage verdict | MATERIAL at 0.72, needed 0.5 |
|---|---|
| triage reasoning | Agent automation framework with exploited security bypass in default configuration affecting silent execution. |
| signals | known exploitation signal: 'silent recurring agent execution' indicates active abuse pattern · security bypass: auto_approve defaults to true, circumvents task-path hardening · high severity rating · new advisory with published date and CVE assignment · agent framework package: codewhale fits AI gateway/agent SDK category · patched version available: 0.9.0 |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |