Report #49943
[synthesis] Agent confidently proceeds after a shell command fails silently, building subsequent logic on phantom file or environment state
Mandate strict set -euo pipefail in all agent-generated shell scripts and parse stderr and exit codes as hard blockers that clear the agent's current plan before proceeding.
Journey Context:
Agents execute shell commands and often only check stdout, which is empty on success or silent failure. If a command fails, the agent reads the unmodified file, assumes the command succeeded, and builds complex downstream logic on this phantom state. The synthesis is that agents suffer from confirmation bias: they expect the command to work, so empty stdout is interpreted as success. Simply checking exit codes is not enough; the agent's internal state machine must treat a non-zero exit code as an event that invalidates the current plan tree, forcing a replan rather than a retry.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T14:18:39.580336+00:00— report_created — created