Agent Plan | Zyeon UI

status='ready' — 7 proposed steps. Step 1 is required (locked toggle), step 4 is high-risk and step 7 starts excluded. Untick a step and the numbering, the ≈24 min / $0.52 estimate and the high-risk warning all recompute for the run that would actually happen. Approve fires once no matter how fast you click it; Request changes sends your note plus the steps you kept.

Migrate billing onto the new pricing table

“Move us onto the new pricing table before Friday's release — without breaking anyone's existing subscription.”

Needs review

Read src/billing/pricing.ts and the current subscriptions schemalow risk
Why: Everything below depends on the column names, so this runs first and can't be dropped.
20 s$0.01
Write a migration that adds price_tier and backfills it from the legacy plan_codemedium risk
Why: A new column keeps the old one readable, so a rollback is a deploy and not a restore.
Backfills 41,800 rows in one statement — the table is locked for the duration.
3 min$0.06
Take a logical backup of subscriptions before touching itlow risk
Why: Cheap insurance: the backfill is the only step here that can't be undone by editing code.
45 s$0.02
Apply the migration to the production databasehigh risk
Why: Staging has 12 rows and none of the legacy plan codes, so it can't prove the backfill works.
Writes to production. If the backfill maps a plan code wrong, 41,800 customers get billed on the wrong tier and the fix is a restore.
8 min$0.12
Update the 14 call sites that still read plan_code, starting with the invoice renderermedium risk
Why: The renderer is the only one customers can see, so it goes first and the rest follow.
12 min$0.31
Run the billing test suite and fix whatever the migration brokelow risk
Why: Unbounded on purpose — nobody can estimate a failure they haven't seen yet.
Post a summary to #billing and email finance@ with the tier mappinghigh risk
Why: Finance reconciles tiers by hand every Monday; they need the mapping before then.
Sends mail outside the company. There is no unsend.
2 min$0.02

no callback fired yet

status='ready' with showRationale={false} — 11 entries in, 10 rows out: the last entry reuses an earlier id and is dropped, because a toggle keyed by id would otherwise move two rows at once. The first step is a 300-character instruction — it wraps inside the card instead of widening it and is never clipped. The footer names the one included step that carries no estimate instead of counting it as free.

Refactor the ingestion pipeline

Needs review

Replace the hand-rolled retry loop in packages/ingest/src/pipeline/consumer/retry-with-exponential-backoff-and-jitter.ts with the shared policy from @acme/resilience, keeping the existing 5-attempt ceiling, the 30-second cap and the dead-letter hand-off, and delete the two now-unused helpers next to itmedium risk
Touches the retry path every consumer shares.
25 min$0.62
connect the cross-platform monitor in /Library/floodlight_failing_father.shtml
30 s$0.02
compress the online feed in /net/incidentally_paralyse.rng
1 min$0.05
navigate the solid state port in /private/var/accelerator_whoa_intellect.aachigh risk
Rewrites files outside the repository root.
2 min$0.08
bypass the bluetooth pixel in /etc/namedb/dearly.mar
3 min$0.11
parse the bluetooth firewall in /boot/defaults/victoriously.tiff
4 min$0.14
hack the auxiliary program in /usr/ports/evince_metal.dump
4 min$0.17
calculate the virtual hard drive in /System/onto_configuration_even.sh
5 min$0.20
quantify the haptic microchip in /home/yogurt_brief_um.dmg
index the online sensor in /usr/local/bin/fully_yahoo.msp

readOnly — the same plan replayed in a transcript. Every toggle keeps its state and its aria-checked, but nothing can be changed and no decision is offered: a plan you can still edit after it was answered is not the plan that ran.

Migrate billing onto the new pricing table (approved 2 days ago)

“Move us onto the new pricing table before Friday's release — without breaking anyone's existing subscription.”

Read only

Read src/billing/pricing.ts and the current subscriptions schemalow risk
Why: Everything below depends on the column names, so this runs first and can't be dropped.
20 s$0.01
Write a migration that adds price_tier and backfills it from the legacy plan_codemedium risk
Why: A new column keeps the old one readable, so a rollback is a deploy and not a restore.
Backfills 41,800 rows in one statement — the table is locked for the duration.
3 min$0.06
Take a logical backup of subscriptions before touching itlow risk
Why: Cheap insurance: the backfill is the only step here that can't be undone by editing code.
45 s$0.02
Apply the migration to the production databasehigh risk
Why: Staging has 12 rows and none of the legacy plan codes, so it can't prove the backfill works.
Writes to production. If the backfill maps a plan code wrong, 41,800 customers get billed on the wrong tier and the fix is a restore.
8 min$0.12
Update the 14 call sites that still read plan_code, starting with the invoice renderermedium risk
Why: The renderer is the only one customers can see, so it goes first and the rest follow.
12 min$0.31
Run the billing test suite and fix whatever the migration brokelow risk
Why: Unbounded on purpose — nobody can estimate a failure they haven't seen yet.
Post a summary to #billing and email finance@ with the tier mappinghigh risk
Why: Finance reconciles tiers by hand every Monday; they need the mapping before then.
Sends mail outside the company. There is no unsend.
2 min$0.02

A decision whose write fails. onApprove and onRequestChanges return a promise that rejects after 900 ms: the badge reads 'Sending…' and nothing claims the plan went to the agent; then the card rolls back — draft intact, gate re-armed, and an alert says why — so you can decide again.

Migrate billing onto the new pricing table

“Move us onto the new pricing table before Friday's release — without breaking anyone's existing subscription.”

Needs review

Read src/billing/pricing.ts and the current subscriptions schemalow risk
Why: Everything below depends on the column names, so this runs first and can't be dropped.
20 s$0.01
Write a migration that adds price_tier and backfills it from the legacy plan_codemedium risk
Why: A new column keeps the old one readable, so a rollback is a deploy and not a restore.
Backfills 41,800 rows in one statement — the table is locked for the duration.
3 min$0.06
Take a logical backup of subscriptions before touching itlow risk
Why: Cheap insurance: the backfill is the only step here that can't be undone by editing code.
45 s$0.02
Apply the migration to the production databasehigh risk
Why: Staging has 12 rows and none of the legacy plan codes, so it can't prove the backfill works.
Writes to production. If the backfill maps a plan code wrong, 41,800 customers get billed on the wrong tier and the fix is a restore.
8 min$0.12
Update the 14 call sites that still read plan_code, starting with the invoice renderermedium risk
Why: The renderer is the only one customers can see, so it goes first and the rest follow.
12 min$0.31
Run the billing test suite and fix whatever the migration brokelow risk
Why: Unbounded on purpose — nobody can estimate a failure they haven't seen yet.
Post a summary to #billing and email finance@ with the tier mappinghigh risk
Why: Finance reconciles tiers by hand every Monday; they need the mapping before then.
Sends mail outside the company. There is no unsend.
2 min$0.02

status='loading' — the proposal itself hasn't arrived.

status='empty' — the agent answered without proposing anything to run. A proposal that arrives with zero steps renders this too.

No plan to review yet.

The agent will propose its steps here before it runs any of them.

status='error' — the proposal failed to load. A different failure from a plan you disagree with, so it reads differently and offers a retry: pressing it swaps in the ready payload.

Couldn't load this plan

The planning request timed out after 30s. request-id: 9f14-2ac7