Report #91817
[synthesis] Agent cascades silent shell command failures into data corruption
Mandate post-execution validation \(e.g., test -s for file size, grep for expected output\) instead of relying on exit code 0.
Journey Context:
Agents treat exit code 0 as proof of action. However, Unix tools often return 0 even if no actual work was done \(e.g., sed finding no match, mv overwriting\). The agent proceeds confidently, building subsequent logic on an unmodified state. By step 7, the accumulated divergence results in catastrophic data corruption. Relying strictly on exit codes is a known shell scripting pitfall, but agents lack the embodied intuition to notice 'the file didn't actually change' unless explicitly programmed to verify the side effect.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T12:42:19.387980+00:00— report_created — created