fix(cron): validate bot-chat deliver/failure_deliver on dashboard update

[nftpoetrist]

Summary

The CLI and the cronjob tool both run every create/update through tools.cronjob_tools._validate_bot_chat_deliver, which rejects a bot-chat[:<profile>] deliver/failure_deliver target whose named profile doesn't exist on this machine — at write time, not only when a run (or a failure) later needs to route through it and finds nothing there.

The dashboard's PUT /api/cron/jobs/{job_id} never went through that check for either field. It calls cron.jobs.update_job (via _mutate_cron_for_profile) directly after only text-normalizing the values in _normalize_dashboard_cron_updates.

Is this a new regression or pre-existing? Pre-existing, and it predates failure_deliver/NS-788. _validate_bot_chat_deliver was introduced in a2da0ab797 (the original bot-chat delivery feature) for the CLI/tool create+update paths only — the dashboard's update lane never called it back then either, so deliver has been unvalidated on the dashboard since bot-chat delivery shipped. failure_deliver simply inherited the same gap when NS-788 added it.

Today's same-day follow-up commit ("validate failure_deliver at preflight and dashboard update lanes", #101373) did not close this — I checked its diff to hermes_cli/web_server.py directly: it only added failure_deliver to the dashboard's text normalizer (empty clears the override instead of coalescing to a target, matching deliver's normalization). It did not add the bot-chat profile-existence check. And _preflight_check_delivery explicitly skips bot-chat targets for both deliver and failure_deliver ("Unknown-profile failures surface per run in last_delivery_error ... and are validated at create time") — deferring entirely to the create/update-time validator. So _validate_bot_chat_deliver was the only safety net for bot-chat targets, and the dashboard never called it.

Fix

Wire _validate_bot_chat_deliver into _update_cron_job_sync via a new _validate_dashboard_cron_bot_chat_deliver helper, called right after normalization and before the update is persisted (same point the CLI/tool path validates relative to its own update). Covers both deliver and failure_deliver.

Scope note: the dashboard's CREATE endpoint (CronJobCreate) does not expose failure_deliver at all, so only the update path needed a failure_deliver fix. CronJobCreate does expose deliver and has the identical pre-existing gap on create (also never calls _validate_bot_chat_deliver) — that's out of scope for this PR, which is focused on the update lane the same-day follow-up work targeted; happy to open a follow-up for create if wanted.

Empirical verification

Before the fix, calling update_cron_job with failure_deliver="bot-chat:definitely-not-a-real-profile" persisted the value with no exception raised anywhere in the path.

Added two regression tests in tests/hermes_cli/test_web_server_cron_profiles.py:

• test_update_cron_job_rejects_unknown_bot_chat_failure_deliver
• test_update_cron_job_rejects_unknown_bot_chat_deliver

Both assert: HTTPException 400 with the same "not found on this gateway's machine" message the CLI path raises, the scheduler provider is not notified, and the job's stored field is left unchanged.

Mutation-verified via patch-file diff/revert/reapply (not git stash — this worktree shares a .git with several sibling worktrees): reverted only hermes_cli/web_server.py, confirmed both new tests fail with DID NOT RAISE HTTPException, reapplied the patch, confirmed both pass.

Test plan

• tests/hermes_cli/test_web_server_cron_profiles.py — 33 passed (full file, includes the 2 new tests)
• tests/cron/test_cron_failure_deliver.py — 24 passed
• tests/tools/test_cronjob_tools.py + tests/hermes_cli/test_cron_dashboard_off_loop.py + tests/hermes_cli/test_cron_fire_dashboard.py — 100 passed
• Manually confirmed valid updates (deliver: local, failure_deliver: "" clear, bare bot-chat, bot-chat:<existing-profile>) still pass through unaffected

Competitor check

Searched gh pr list --search for "dashboard cron failure_deliver", "cron dashboard bot-chat validate", "_validate_bot_chat_deliver dashboard", "cron dashboard deliver validation", "bot-chat deliver dashboard", "cron dashboard bot-chat profile", "cron update_job bot-chat validate", "dashboard cron deliver bot-chat profile not found" (state=all). No open or merged PR addresses this specific gap; #101373 (merged today) is the closest hit but only did text normalization as shown above.

[Enough1122]

AI code review — automated review for reference; please use your judgment.

PR #102091 — fix(cron): validate bot-chat deliver/failure_deliver on dashboard updates

• hermes_cli/web_server.py:13041-13068 adds _validate_dashboard_cron_bot_chat_deliver() reusing tools.cronjob_tools._validate_bot_chat_deliver for both deliver and failure_deliver, raising 400 on an unresolvable bot-chat profile; wired into _update_cron_job_sync (hermes_cli/web_server.py:13438-13444) next to the existing context_from validation. Closes a pre-existing gap both fields had: CLI and cronjob tool fail loud up front, but dashboard PUTs only surfaced it per-run as last_delivery_error.
• Only present fields are checked (if field not in updates: continue), so partial updates not touching delivery are unaffected.
• Tests tests/hermes_cli/test_web_server_cron_profiles.py:1098-1183 pin 400s for unknown profiles in both fields.

Non-blocking:

• The import is function-local (consistent with this module's heavy-lazy-import style); no cycle risk either way since tools.cronjob_tools doesn't import the web server.

Verdict: LGTM.