Observed (prod, 2026-06-04 ~21:04)
hn-monitor was redeployed ~20:30 via CLI with --on-exists update. The original wizard deployment carried SLACK_CHANNEL in its inputValues (set via picker). The CLI update did not re-supply it, and the update silently dropped the stored input. Next tick:
ctx.llm configured from sandbox credentials provider:anthropic source:CLAUDE_CODE_OAUTH_TOKEN model:claude-haiku-4-5-20251001
Required input 'SLACK_CHANNEL' has no value
(exit 1 — the run died after the credential chain succeeded.)
Expected
deploy --on-exists update should inherit the prior deployment's inputValues unless explicitly overridden by --input KEY=value. Silently dropping inputs on update is a footgun class — this is its first observed instance; any persona with required inputs redeployed via CLI update will hit it.
Fix sketch
- On update, merge: prior deployment inputValues ← CLI-provided
--input overrides.
- An explicit empty/unset should require a deliberate flag (e.g.
--clear-input KEY), not be the default.
- Test: deploy with input via wizard-equivalent, update without
--input, assert input survives.
Context: observed by claude-4 during the overnight 2026-06-04 run; immediate workaround is one redeploy with --input SLACK_CHANNEL=<channel-id>.
Observed (prod, 2026-06-04 ~21:04)
hn-monitor was redeployed ~20:30 via CLI with
--on-exists update. The original wizard deployment carriedSLACK_CHANNELin its inputValues (set via picker). The CLI update did not re-supply it, and the update silently dropped the stored input. Next tick:(exit 1 — the run died after the credential chain succeeded.)
Expected
deploy --on-exists updateshould inherit the prior deployment's inputValues unless explicitly overridden by--input KEY=value. Silently dropping inputs on update is a footgun class — this is its first observed instance; any persona with required inputs redeployed via CLI update will hit it.Fix sketch
--inputoverrides.--clear-input KEY), not be the default.--input, assert input survives.Context: observed by claude-4 during the overnight 2026-06-04 run; immediate workaround is one redeploy with
--input SLACK_CHANNEL=<channel-id>.