Missive · focused repair plan

Verification and Constraints

READY FOR REVIEW

Team space = sales stage. Unread = useful action. Blue note = why the action matters.

Preparation only · no cleanup launchedSeptember 24, 2026

Goal and finish line

End goal: Make Missive show the right conversations for attention, with readable notes and context-aware follow-ups.

Done means: The selected conversations are correctly filed, lost leads stop receiving chase drafts, useful work stays unread, and future automation follows the agreed rules.

0 / 17execution criteria verified

6deliverables

0product decisions open

Your settled choices

Blue Comic Sans notes stay above the email draft. Normal case, short sections.
New replies preserve owned M/G stages and Active Clients.
Confirmed losses: owner's Lost stage, read, inappropriate automation drafts backed up and removed.
No client email sends. Preserve human drafts and Snooze.

What the inspection found

Confirmed: the current code explicitly uppercases the review header.
Confirmed: a full call transcript contains a clear rejection, while the automation's context was missing that call.
Confirmed: Lost-stage conversations still appear with automation drafts.
Confirmed: team names changed to Positive replies, Negative replies and Active Clients. Parts of the code still rely on old names.
Unverified: Comic Sans preservation in the actual Missive composer. The executor must test it, not just generate styled HTML.

This is a repair of the existing system. No new CRM, new team spaces or extra scheduled jobs.

How the notes should look

Illustrative layout only. This is not a screenshot or proof that Missive preserves the font.

Helyzet
Az érdeklődő visszajelzése alapján egyeztetés szükséges.

Legutóbbi megállapodás
A korábban felajánlott időpontok már elmúltak.

Következő lépés
Friss időpontot egyeztetni, ha a hívás még nem történt meg.

Források
Az ellenőrzött levelek és a kapcsolódó hívás itt hivatkozhatók.

Belső jegyzet. Küldés előtt töröld ezt a részt.

Kedves …,

Ide kerül az ellenőrzött, ügyfélnek szánt levél, normál betűtípussal.

Follow-up rule, without guesswork

Allowed stage + not snoozed + no draft + due and justified by context → one unsent draft, marked unread.

Allowed: Active Clients; M/G Contacted, no call booked; M/G Proposal sent; M/G Call booked; or genuinely no team. Successful drafting in no team → M - Contacted, no call booked.

Fresh queues, Lost, Paid, FUP later, Other, Updates and intake buckets do not qualify. No-team does not mean permission to chase arbitrary personal mail.

Deliverable flow

Selected conversations

Freeze IDs and preserve before-state.

Email + call evidence

Decide what action is warranted.

Rules + cleanup

Repair the cause and old drafts.

Verified Missive state

Prove actual behavior and rerun safety.

Parent ↔ executor

Parent decision

The three grilling questions are settled. Review this plan, then send Launch to start one Sol Medium executor.

Executor status

No executor has been created. Live progress and evidence will appear here in blue after launch.

Work and proof

Each checkbox needs evidence

1. Reconcile the attention queue

classify actual work, not merely reduce unread count.

0 / 3

Criteria, constraints and proof

When the screenshot examples and opened conversation are resolved using the current provider inbox, then a private manifest appears with cutoff, conversation IDs, exact scoped count, lookup/pagination evidence and before-state. The mutation scope is these examples only, including any now read: the eleven named conversations in the private source manifest. Resolve identities from provider, never names alone. Screenshot is source evidence, not a current unread census. Current unread enumeration is lookup evidence only, not authority to alter unrelated inbox conversations. Unresolved names require exact email/thread evidence before inclusion. When each scoped conversation is assessed against complete relevant email and available call context, then the manifest records Keep/move/mark-read/retain-unread/draft-repair with reason and evidence. All rows have a verified result or an explicit unresolved dependency, not a blanket Done. Audit existing automation drafts separately from future follow-up eligibility: a fixed whitelist does not repair a previously saved inappropriate draft. For Lost/FUP later examples, evaluate the draft against full current evidence and preserve human content; do not infer that a team label alone authorizes deletion. A future follow-up agreement must retain its actual date and Snooze. When this Launch-authorized cleanup is applied, then useful reply or draft work remains unread in the correct stage and no-action conversations appear in their appropriate existing space, not automatically archived. Read back stage and draft state and capture actual post-change Missive screenshots for attention changes.

Preserve these constraints

preserve human drafts, existing ownership and Snooze; no bulk archive just to clear counts.
Do not assume an old unread is actionable or that every thread with no team is a sales opportunity.
Out-of-scope conversations remain unchanged..

Pending execution: attach provider readback, regression output and exact result here.

Actual Missive after-test screenshot → blind Luna High description → comparison with criterion. No screenshot is verified yet.

2. Respect lost-call outcomes

Prove the requested behavior in the actual system.

0 / 2

Criteria, constraints and proof

When the confirmed-lost call fixture is evaluated with the Sep16 sales-call transcript and any newer evidence, then the private decision records rejection of the proposed spend, owner M and no agreed follow-up date. Source timestamps 21:20, 21:39,22:08. Matt explicitly confirms this lead is lost. When cleanup runs for that confirmed loss, then the actual Missive conversation appears in M - Lost and read, inappropriate automation-owned drafts have recoverable backups and are removed, and future follow-up evaluation produces no draft. Screenshot plus provider draft readback plus regression result prove it.

Preserve these constraints

no claim he cannot afford other services; record declined proposed spend.
A dated future request is FUP later plus Snooze, not Lost.
Preserve human drafts and flag uncertain draft ownership.
A later verified agreement outranks an older refusal..

Pending execution: attach provider readback, regression output and exact result here.

Actual Missive after-test screenshot → blind Luna High description → comparison with criterion. No screenshot is verified yet.

3. Make draft notes readable

Prove the requested behavior in the actual system.

0 / 3

Criteria, constraints and proof

When a new reply/follow-up draft is generated, then the Missive composer shows notes ABOVE the email, in blue (#1559c9) Comic Sans, normal case, with four concise Hungarian sections: Helyzet, Legutóbbi megállapodás, Következő lépés, Források. Show factual context and actionable missing evidence, not raw errors/file lists. Client text stays normal font/color. Actual screenshot is required to prove provider rendering; do not infer font from HTML alone. When an existing automation draft in scope has the old INTERNAL REVIEW uppercase header, then its saved header is replaced with the same structured format while its recipient, sender, subject, client-facing body and references remain unchanged unless independently found inappropriate under the loss rule. Before/after comparison and provider readback prove preservation. When the operator reviews a completed draft, then the note visibly says in normal case to remove it before sending. No send integration is added; the automation never sends. If Missive strips Comic Sans, demonstrate actual behavior and report this exact unsatisfied style criterion rather than claiming pass.

Preserve these constraints

notes remain above the draft per Matt's explicit choice, not moved to separate comments.
No fabricated call coverage and no full-capital blocks.
No overwriting human edits, rerun idempotency required..

Pending execution: attach provider readback, regression output and exact result here.

Actual Missive after-test screenshot → blind Luna High description → comparison with criterion. No screenshot is verified yet.

4. Restrict automatic follow-ups

Prove the requested behavior in the actual system.

0 / 3

Criteria, constraints and proof

When a due, context-justified follow-up is evaluated in Active Clients, M/G - Contacted, no call booked, M/G - Proposal sent, M/G - Call booked, or genuinely no team, with no Snooze and no existing draft, then at most one unsent follow-up appears and the thread is unread. Team eligibility is necessary, not sufficient: preserve due dates, human context and opt-out/loss protections. When the same input has Snooze, any existing draft, or any other team including Positive replies, Negative replies, Lost, Paid, FUP later, Other, Updates or Instantly Outreach, then no follow-up draft is created. Test all stages and race changes immediately before writing. When a follow-up is successfully saved for a no-team conversation, then it appears in M - Contacted, no call booked. Failed draft creation must not move it; failure after creation must retry only the move without duplicating the draft. Provider readback and injected-failure tests prove both.

Preserve these constraints

do not alter reply drafting into a follow-up loophole; no cold outreach to arbitrary unfiled personal mail.
No follow-up from a missing transcript when it could change deal status.
Retrieve available sources or flag one concise unresolved action, rather than write an uninformed chase..

Pending execution: attach provider readback, regression output and exact result here.

Actual Missive after-test screenshot → blind Luna High description → comparison with criterion. No screenshot is verified yet.

5. Route incoming replies by identity

Prove the requested behavior in the actual system.

0 / 3

Criteria, constraints and proof

When a new unclaimed incoming reply is matched by exact Message-ID to Instantly and interested, then it appears in Positive replies and unread. When the matched reply is human and not interested, then Negative replies and unread. Automated replies do not enter either human queue. Readbacks plus actual screenshots and controlled fixtures prove each branch. When an incoming reply belongs to an owned M/G stage or Active Clients, then its ownership, stage and Snooze state stay unchanged and the new human reply becomes unread. Snooze still prevents follow-up creation. Read back any provider-triggered Snooze change and record it rather than silently applying a different policy. Do not auto-reopen a lost sales chase merely because a new email exists; retain new human attention without creating an ineligible follow-up. When the team display names are renamed while IDs stay the same, then eligibility/routing outcomes remain unchanged. Tests use verified IDs, not stale Fresh +/Fresh -/Clients string keys.

Preserve these constraints

preserve existing positive-only Slack body/link cards and deduplication.
Keep bounce auto-archive protection and shared API pacing from the previous repair.
Do not re-route an old owned conversation simply because its original source was Instantly..

Pending execution: attach provider readback, regression output and exact result here.

Actual Missive after-test screenshot → blind Luna High description → comparison with criterion. No screenshot is verified yet.

6. Prove and hand off the repair

Prove the requested behavior in the actual system.

0 / 3

Criteria, constraints and proof

When regression tests and safe live provider cases run, then evidence covers every above branch, a deployed-source hash match and at least one ordinary cycle from the verified existing production schedule. Record that schedule and cycle identity; if no production schedule exists, run a controlled invocation and leave the scheduled-cycle criterion explicitly unsatisfied without creating a schedule. No client email is sent to manufacture evidence. Clearly distinguish controlled replay from a naturally arriving reply. When the final board is reviewed, then every passed GUI criterion has a real after screenshot, independent gpt-6-luna high description supplied without expected result, and parent/executor comparison. Maintain a criterion-to-evidence matrix with object IDs, timestamp, source revision, real-vs-fixture label, screenshot/private-original mapping and blind-description model/session. All other criteria link to provider readbacks/tests, including sent-state checks, applicable Slack dedup/body/link and bounce evidence, and public URL revision/redaction readback. Do not generate client messages or Slack invitations for testing. Unsatisfied criteria remain unchecked, with exact resume instructions. When an unchanged scoped conversation is reprocessed, then no duplicate draft, duplicate Slack alert, repeated attention note or unnecessary unread reset appears. The final manifest accounts for all scope rows and reports remaining exceptions and action counts.

Preserve these constraints

no green checks from worker claims alone; no widening scope to old handoff/skills projects.
HTML stays at one stable bare URL and is updated live.
Public board uses sanitized examples and counts; private emails/transcripts/tokens and unredacted screenshots stay in local evidence, with no new access gate introduced.
Publish only redacted screenshot derivatives that retain the tested visual state, and keep their private-original mapping locally..

Pending execution: attach provider readback, regression output and exact result here.

Actual Missive after-test screenshot → blind Luna High description → comparison with criterion. No screenshot is verified yet.

Inputs and review linksExact local sources stay private

Private goal and source manifest live in the task's missive-tidy-2026-09-24 handoff folder. They contain exact conversation IDs and the recovered call transcript location. Public board contains no private email bodies.

Open the scheduling example in Missive · Open the confirmed-lost example in Missive

Open decisions and technical checks

No further business decision needed. The executor must resolve exact current scope identities, verify access to call evidence, and test Missive's font rendering. Missing evidence stays visible rather than being filled with invented context.

Recovery state

Not started. Before changes: snapshot conversations and drafts, verify runtime ownership, retain recovery paths. If blocked, leave the precise affected criterion unchecked and preserve the resume point.

Current parent action: independent plan critique completed; review this board, then send your explicit Launch.