Review what exists, see the usability gaps, and approve a precise finish line.
Owner: parent · Prepare onlyUpdated: 24 September 2026Executor: not launched
Goal and finish line
End goal: Give Gergő a portable, reliable sales workflow he can run from a new machine, from a fresh reply to a reviewed offer and signing link.
Done means: One private starting page, one versioned ZIP, three clear commands, a complete worked example, and evidence for the actual intake and commercial handoffs. Remaining account/device dependencies are visible and owned.
0 / 24future criteria verified
6final deliverables
3primary sales skills
Rules that stay fixed
Preserve the original skills. Package the three existing kit entry points without GitHub.
Positive incoming replies only in Slack. Manual claim and G/M move. Drafts requiring attention stay unread.
Matt invites Gergő to Slack. No client sends, signatures or new paid seat during finalization tests.
Proof standard: exact output, provider readback and a fresh operator rehearsal. A unit test or finished agent is insufficient by itself.
Keep the core. Finish the handoff.
Keep
Three entry skills. Shared Missive login. Positive-only Slack alerts. Manual claim and G/M stages. Review HTML before signing.
Fix before acceptance
Unproved receiving-machine setup, incomplete full preparation, internal notes inside draft bodies, and conflicting release/status information.
Track separately
The historical follow-up scan and renewable Wispr access. They must remain visible, but a collector should not keep the whole handoff in a polling loop.
Evidence boundary: live public guides and private Notion text were read on 24 September. ZIP bytes, skill files and saved acceptance reports were inspected. Missive, Slack and cloud runtime behavior below comes from dated provider receipts, not a new live system test. A reported pass is not a fresh re-test.
Confirmed artifact or settled requirementRecommendation / inferenceNot yet proved / human dependency
Review the work already created
Open an artifact if needed. Each card says what to retain, what is insufficient, why it matters, and the smallest useful correction. Private links require your existing access. Local files are not public downloads.
Gap: It presents the full preparation route as the expected output, without a prominent current readiness label. Figma and a complete fresh preparation run are still unproved.
Why it matters, recommended fix and evidence
Keep: Three visible skills, copyable prompts, clear contract review/finalize split. Keep this structure.
Why it matters: A new operator can reasonably expect the advertised route to work immediately and discover access or missing outputs during a call.
Recommended correction: Keep the three commands. Add setup readiness and one complete worked example. Make Notion the employee starting point, with this page as the detailed skill guide.
Evidence: Live page text + TEST-RESULTS.md explicitly exclude full 14-asset, Figma and receiving-machine acceptance.
Gap: Step 5 tells the operator to remove an ALL-CAPS internal header from the outgoing draft. Screenshots sit in a separate gallery. The page and public guide repeat instructions.
Why it matters, recommended fix and evidence
Keep: Real annotated screenshots, ZIP attachment, Slack claim, manual G/M move, unread and Snooze instructions.
Why it matters: Forgetting to delete the header can expose internal notes to a lead. Repeated instructions can drift, and a detached gallery makes actions harder to follow.
Recommended correction: Put internal context in a supported private comment or local review note. Keep draft bodies client-ready. Put one useful screenshot beside each action, with one canonical setup/download section.
Evidence: Notion fetched content: “töröld a teljes belső fejlécet”; screenshot captions also tell the user to delete it.
BYTES VERIFIED · local
Portable ZIP v3
Open existing artifactLocal file · copy this path into Codex: /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/clientsflow-sales-kit-onboarding-v3.zip
Gap: Clean/repeat installs were tested on the builder Mac. The receiving OS, own-machine access and full run remain unverified.
Why it matters, recommended fix and evidence
Keep: 527-file manifest, preserved original skills, sanitized examples, three entry commands. SHA-256 fd695af25f67636223c141faf4bc2157ab0225576694636cb7e533b49067dbaf.
Why it matters: Portability of files does not prove that Gemini, publishing, Figma, Missive and DocuSeal work on Gergő’s computer.
Recommended correction: Repair only demonstrated setup failures, build one final version, and provide an automated readiness result plus a receiving-machine receipt.
Evidence: TEST-RESULTS.md; parent-package-v3-acceptance.json; current archive hash matches the accepted receipt.
PREPARED · not sent
Installation prompt and handoff message
Open existing artifactLocal file · copy this path into Codex: /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/KEZDD-ITT.md
Gap: The installation request bundles many services and a full rehearsal into one paragraph. It does not provide the employee with a compact ready / action-needed result for each stage.
Why it matters, recommended fix and evidence
Keep: Codex installs without GitHub. Existing skills are preserved. The separate UZENET-GERGONEK.md explains shared login and Matt’s Slack invitation.
Why it matters: A failed optional connection can look like total failure, leaving the operator to diagnose dependencies.
Recommended correction: Keep one pasteable prompt. Make it produce a short setup report, exact missing sign-ins, and an explicit resume command. Update the unsent message to reference the exact final archive.
Evidence: KEZDD-ITT.md and UZENET-GERGONEK.md read directly on 24 Sep.
Gap: A 7/8 counter combines a completed historical sort, synthetic runtime checks and an unfinished all-history follow-up bootstrap. Its seed count is 8,116, while later collector reports differ.
Why it matters, recommended fix and evidence
Keep: Preserves deployed runtime, migration, rollback and explicit incomplete states.
Why it matters: The counter does not answer “Can Gergő work independently?” Historical scanning dominates attention even though it is a separate background job.
Recommended correction: Keep this as an engineering receipt. This finalization board becomes Matt’s active review surface. Split operator readiness, runtime proof and background cleanup, with dated evidence.
Evidence: Live board readback 24 Sep + collector report. No live cloud refresh performed for this plan.
Gap: The corrected card is an internal fixture. A natural positive human reply has not been observed through the entire route. Gergő’s notifications are unverified.
Why it matters, recommended fix and evidence
Keep: New card shows incoming body and Missive/verified website links, with no reply draft or buttons. The old broad producer was stopped.
Why it matters: Formatter tests prove card content, but not arrival, classification, delivery and human claim on the employee’s device.
Recommended correction: Use a controlled external test through the real intake path, plus no-alert negative/OOO/opt-out cases. Record synthetic versus natural evidence separately. Verify device delivery when Gergő is available.
Evidence: collector-final.md and B-final.md. Existing message URI recorded in the report, not reopened in this review.
BOUNDED AUDIT DONE · corpus incomplete
Historical template audit
Open existing artifactLocal file · copy this path into Codex: /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/historical-ten-chain-audit.md
Gap: This does not meet the earlier request for ten complete linked journeys. Conversely, it is not evidence that the current templates are bad or should be replaced.
Why it matters, recommended fix and evidence
Keep: Ten distinct deals were checked. Four have all four artifact types; two also have a verified sent offer and matching DocuSeal submission. Real conflicts support case-specific terms.
Why it matters: Chasing an arbitrary complete corpus could postpone handoff; claiming ten complete cases would overstate the evidence.
Recommended correction: Retain traceable templates and add a usable selection guide with worked examples. Account for all ten requested cases and missing joins. Do not assign this research to Gergő or invent missing artifacts.
Evidence: Audit’s opening result and installed/shipped writing-structures.md and template-selection.md inspected.
PARTIAL ACCEPTANCE
Sales-kit acceptance limits
Open existing artifactLocal file · copy this path into Codex: /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-kit/TEST-RESULTS.md
Gap: It correctly says ten complete historical chains were not audited, but does not link the later bounded ten-deal findings. It also explicitly excludes a fresh full prep, native Windows run and Figma import.
Why it matters, recommended fix and evidence
Keep: Installer, Gemini A/B, commercial checks and a test DocuSeal/Missive round trip have real evidence.
Why it matters: The limitation is honest. The missing cross-reference still leaves a reviewer to reconcile the later audit and release evidence manually.
Recommended correction: One release manifest must point to current evidence, name exactly what passed and retain every untested requirement.
Evidence: TEST-RESULTS.md compared with the newer historical audit and parent-final-acceptance.md.
Gap: The three-month sorting is done; the all-history follow-up seed is a different unfinished process. The Codex collector reported blocked while the cloud process still ran. Wispr’s renewable grant remains absent.
Why it matters, recommended fix and evidence
Keep: 108 reported tests, nine guards, editable unsent/unread drafts, manual placement and Snooze protection.
Why it matters: “Collector blocked” can sound like sales intake is broken. Repeated status polling produces activity without moving employee readiness forward.
Recommended correction: Retire routine agent polling after an operational handoff. Record the background job owner, stop/resume method and completion condition. Keep full cloud transcript grounding as a separately visible pending requirement.
Evidence: collector-final.md, B-wispr-authorization.md, and latest terminal collector message (8,702, incomplete, zero retries). Those values are historical snapshots, not current live state.
Gap: The delivered set has 17 spaces, including Clients, Other and Updates. The original request listed 14 and asked that only those be visible. Later authority for the extra visible spaces needs to be traced before calling this a defect.
Why it matters, recommended fix and evidence
Keep: Fresh + and Fresh - plus separate G/M stages were created, with rollback evidence.
Why it matters: Extra queues make the intended daily workspace less obvious. Deleting spaces merely to simplify the sidebar could lose useful placement.
Recommended correction: Check later confirmed instructions. Preserve spaces and data; if the extra sidebar entries were not approved, hide them reversibly so only the requested sales stages remain visible. Document any justified exception.
Evidence: Original stage list in this conversation, A-final.md and private Notion current stage list.
PREPARED · not sent
Exact unsent employee handoff message
Open existing artifactLocal file · copy this path into Codex: /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/UZENET-GERGONEK.md
Gap: The final version and readiness are not yet reconciled with the final package.
Why it matters, recommended fix and evidence
Keep: One message linking the ZIP and setup Markdown, with Matt responsible for the manual Slack invitation.
Why it matters: A correct guide does not prevent an old message from distributing the wrong archive or overstating readiness.
Recommended correction: D5 updates this message with the exact final archive and setup links and verified sender [email protected]. Keep it unsent.
Open existing artifactLocal file · copy this path into Codex: /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/parent-final-acceptance.md
Gap: Past acceptance cannot prove the future final package, full preparation or receiving-device behavior.
Why it matters, recommended fix and evidence
Keep: Explicit prior acceptance and unresolved boundaries.
Why it matters: A single prior green result can conceal unresolved cross-system requirements.
Recommended correction: D6 links current evidence per requirement and retains all historical limitations until disproved.
Collector reportLocal file · copy this path into Codex: /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/collector-final.md
sales-prep-kitLocal file · copy this path into Codex: /Users/agency/.agents/skills/sales-prep-kit/SKILL.md
offer-in-email-kitLocal file · copy this path into Codex: /Users/agency/.agents/skills/offer-in-email-kit/SKILL.md
contract-to-docuseal-kitLocal file · copy this path into Codex: /Users/agency/.agents/skills/contract-to-docuseal-kit/SKILL.md
sales-kit-runtimeLocal file · copy this path into Codex: /Users/agency/.agents/skills/sales-kit-runtime/SKILL.md
Public page: critique and plan only. Raw email bodies, contracts, transcript text, protected screenshots and credentials remain in private sources.
End goal system and routeTwo maps · expand if useful
End goal system
Green means the workflow is explicitly requested, not that every step is already verified. Offer and contract review are separate gates.
flowchart TD
IN["<b>Incoming reply</b><br/>Existing intake"] --> Q{"Interested human reply?"}
Q -->|Yes| POS["<b>Fresh + and Slack</b><br/>Body and useful links only"]
Q -->|Other human| NEG["<b>Fresh -</b><br/>No Slack alert"]
Q -->|Automation or spam| NOISE["<b>Noise</b><br/>No sales alert"]
POS --> CLAIM["<b>Claim and move</b><br/>Checkmark then G or M stage"]
NEG --> CLAIM
CLAIM --> PREP["<b>Prepare and call</b><br/>Three files, HTML, two designs"]
PREP --> ACCEPT{"Accepted on call?"}
ACCEPT -->|No| OFFER["<b>Exact offer</b><br/>Review and send intentionally"]
OFFER --> DECIDE{"Accepted later?"}
DECIDE -->|Later| SNOOZE["<b>Follow up later</b><br/>Dated Snooze"]
SNOOZE --> OFFER
DECIDE -->|Yes| CONTRACT["<b>Review contract</b><br/>Saved HTML edits"]
ACCEPT -->|Yes| CONTRACT
CONTRACT --> EDIT{"Edits resolved?"}
EDIT -->|Revise| CONTRACT
EDIT -->|Yes| SIGN["<b>Signing handoff</b><br/>DocuSeal URL and Missive draft"]
class IN,Q,POS,NEG,NOISE,CLAIM,PREP,ACCEPT,OFFER,DECIDE,SNOOZE,CONTRACT,EDIT,SIGN confirmed;
classDef confirmed fill:#E8F5E9,stroke:#2E7D32,color:#173B25;
classDef assumption fill:#FCE4EC,stroke:#C2185B,color:#5B1430;
classDef open fill:#fff,stroke:#94A3B8,color:#475569,stroke-dasharray:5 5;
Read the flow as text
Positive human reply → Fresh + and Slack → manual claim → prep and call. Other human replies → Fresh -, without Slack. Accepted on the call → contract. Otherwise → offer → acceptance or dated follow-up. Contract edits loop until resolved → signing URL and draft.
Route to the end state
flowchart LR
PK["<b>Package exists</b><br/>Builder checks passed"] --> VERIFY["<b>Reconcile and repair</b><br/>Keep valid work, fix proved gaps"]
RT["<b>Runtime exists</b><br/>Controlled provider proof"] --> VERIFY
GAP["<b>Device and access gaps</b><br/>Not yet proved"] --> VERIFY
VERIFY --> TRIAL["<b>Run complete rehearsal</b><br/>Setup, prep, commercial and intake"]
TRIAL --> HAND["<b>Publish one handoff</b><br/>Matching ZIP, guide and message"]
HAND --> RESULT["<b>Accept against proof</b><br/>Complete or exact remaining dependency"]
class PK,RT confirmed;
class GAP,RESULT open;
class VERIFY,TRIAL,HAND assumption;
classDef confirmed fill:#E8F5E9,stroke:#2E7D32,color:#173B25;
classDef assumption fill:#FCE4EC,stroke:#C2185B,color:#5B1430;
classDef open fill:#fff,stroke:#94A3B8,color:#475569,stroke-dasharray:5 5;
Diagram rendering failed. The readable workflow and deliverable cards remain available.
Deliverable flow
Direct acceptance can skip the offer, as shown above. Final signing output uses the accepted ledger and resolved review edits. Green pills are outputs, gray fields are inputs.
Contract reviewCaptures final correctionsEditable HTMLReview URL
Signing handoffMakes the agreement signableFinal PDFSigning URLMissive draft
Work and proof
6 final deliverables · implementation not started
Each card defines an observable result. Expand for exact outputs, acceptance, proof and constraints. Checkboxes are evidence markers, not buttons to approve the plan.
D1
Deliver installation that works independently
Turn the existing ZIP into a reproducible setup on the recipient’s actual computer.
Exact paths are also listed in the executor contract below. ~ means the receiving user’s resolved home and actual Desktop.
Verification criteria
Proof: Inventory against sales-prep-kit/references/source-workflow.md, file existence/hashes and source freshness rationale.
Proof: Request manifests and template hashes. Reuse valid outputs to avoid unnecessary API spend. Preserve generated originals.
Proof: Live URL/content readbacks and desktop/mobile section screenshots. Calculator assumptions labeled. Apply Luna visual QA where required.
Proof: Actual Figma URL, node readback, loaded images/editable text and screenshots. Missing auth remains blocked, not imported.
Preserve these constraints
Preserve the original prep workflow, fixed prompt/template bytes and current verified client facts. Do not replace full prep with two pretty homepages.
Do not substitute an invented Figma URL, missing images, unsupported claims or illustrative CPC as observed performance.
Dependencies: D1 core runtime
Required private screenshot proof Rakamaz review index, A/B desktop and mobile sections, and Figma file coverage.
Executor evidence: not yet produced for this finalization run. Add dated receipt links here only after verification.
D3
Prove offers and contracts preserve agreements
Carry one case from full call through an edited contract to the exact signing draft.
Exact paths are also listed in the executor contract below. ~ means the receiving user’s resolved home and actual Desktop.
Verification criteria
Proof: Example-to-source mapping, sanitization check, named required inputs and current-case overrides. Historical audit retains 10 rows and explicit missing joins. Complete recoverable missing joins using permitted sources; preserve missing joins as unmet requirements with exact unavailable evidence and resume steps. Two worked examples and ten candidate rows do not satisfy ten complete chains.
Proof: Field-by-field transcript/ledger/offer comparison, including page count and expiry conditions. No invented deadline, rate, warranty or acceptance.
Proof: Test both entry routes, edit a material clause in review, reload, reconcile ledger and compare final HTML/PDF text and hashes.
Proof: Internal test submission with automatic mail suppressed, provider readbacks, signing URL equality and no duplicate record on resume.
Preserve these constraints
No real prospect commitment, signature or client email during rehearsal. Use isolated internal fixtures for writes and preserve all commercial source facts.
Keep current edited HTML authoritative and existing originals untouched. Ten candidates must never be described as ten complete chains.
Dependencies: D1 provider readiness
Required private screenshot proof Private contract edit after reload, final PDF, DocuSeal fields and unsent Missive draft.
Executor evidence: not yet produced for this finalization run. Add dated receipt links here only after verification.
D4
Prove intake reaches the right person
Verify real routing and remove the manual cleanup trap in outgoing drafts.
Exact paths are also listed in the executor contract below. ~ means the receiving user’s resolved home and actual Desktop.
Verification criteria
Proof: Record received message ID, Instantly classification where applicable, Missive conversation/stage, Slack timestamp and same-thread direct link. Negative, opt-out, OOO, duplicate and manual-stage tests. Keep natural arrival separately unobserved unless witnessed. Trace 17-versus-14 provenance before changing visibility. OOO is an automated reply, not a human Fresh - case; preserve the supported existing handling and prove no positive Slack alert. Identify controlled test sender, recipient and safe test channel before writes.
Proof: Provider payload and screenshot. No generated reply, send controls or guessed website.
Proof: Use supported internal comment or local context note. Read back body, note, sender/thread and unread after create/update/replay. Preserve human drafts and Snooze. Repair generator-produced new/update bodies; do not bulk rewrite existing human-authored drafts. Record compatibility with existing internal context and any remaining legacy cleanup separately.
Proof: Controlled claim rehearsal: inspect existing reaction, add own checkmark, verify G/M placement before composing; if collision, agree one owner. Device receipt after Matt’s invitation. No claim of automatic locking.
Preserve these constraints
Matt invites Gergő to Slack manually. No paid Missive seat, new identity infrastructure or automatic Slack-reaction routing.
No automated client send, no opt-out follow-up, no overwrite of manual stages/human drafts/Snooze. Shared unread is attention, not ownership.
Dependencies: D1 Missive access, Matt Slack invitation for device criterion
Required private screenshot proof Private positive Slack card, exact Missive stage, separate internal context and unread draft.
Executor evidence: not yet produced for this finalization run. Add dated receipt links here only after verification.
D5
Deliver one clear operator handoff
Make the package usable without original chats or an engineering board.
Exact paths are also listed in the executor contract below. ~ means the receiving user’s resolved home and actual Desktop.
Verification criteria
Proof: Read the whole guide as a fresh operator. Put authentic annotated screenshots at relevant steps. Three primary skills only; helpers nested below them.
Proof: Link check and byte/version comparison. Natural Hungarian. Message prepared, not sent. Sender is [email protected], per T0191–T0198 correction.
Proof: Unsupervised trial assignment and final report/gist. No parent coaching, no original chat dependency, no new documentation project.
Proof: Guide review + actual receiving-machine/device receipt when available. Unavailable device proof remains an explicit partial state.
Preserve these constraints
Keep private client screenshots, contracts and credentials out of the public review host. Preserve the existing Notion URL/access and public guide comment identity.
Do not bury the workflow under technical file tables, duplicate skill lists or instructions for Gergő to audit ten clients.
Dependencies: D1–D4 outputs
Required private screenshot proof Private guide’s start, one inline action screenshot and one worked case.
Executor evidence: not yet produced for this finalization run. Add dated receipt links here only after verification.
D6
Close execution with honest operational ownership
Make readiness reviewable without endless collector polling.
Exact paths are also listed in the executor contract below. ~ means the receiving user’s resolved home and actual Desktop.
Verification criteria
Proof: Every D1–D6 checkbox links dated evidence. All checked only with proof; no full COMPLETE while device/Figma/intake criteria remain unproved. DEVICE PENDING and PENDING AUTH are honest terminal partial states for the affected proof, never passed checkboxes or a full-completion claim. Complete every independently achievable artifact before handoff.
Proof: Single dated provider readback and durable job configuration. No completion inferred from a live lease, increasing count or a finished Codex task.
Proof: Reuse B-wispr handoff. If authorized, prove renewable full transcript retrieval and grounded drafting. Otherwise record PENDING AUTH; per-call supplied full transcript supports offers but is not unattended cloud grounding.
Proof: One linked terminal handoff and exact resume points. Event-driven completion notification only if configured and authorized; no new recurring poller by default.
Preserve these constraints
Do not restart, parallelize or cancel a healthy cloud worker just to close a Codex task. Missing grant blocks only the affected capability.
No extra collector between Matt and the single LLL executor. No weakening of original requirements to obtain a green board.
Dependencies: D1–D5 evidence
Executor evidence: not yet produced for this finalization run. Add dated receipt links here only after verification.
What you are reviewing
The proposed scope is repair and acceptance of the existing system. There is one finalizing executor, not another collector layer.
01
Accept the six deliverables as the finish line.
Keep the full prep/Figma promise. Missing access or device proof is a named pending criterion, never a silent scope reduction.
02
Approve moving internal notes out of email bodies.
This is a recommended usability correction, supported by the current Notion instruction to delete a checking header. The exact private-note method is the executor’s choice after proving it works.
03
Keep the historical work visible without blocking daily sales.
Recommendation: do not make archive exhaustion or finding ten complete historical chains a prerequisite for using the proven parts. The executor must preserve the incomplete requirements in its final report. This recommendation does not declare those requirements complete.
Human-dependent steps: Matt’s manual Slack invitation and Gergő’s device sign-in/notification test. Figma may need renewed sign-in. The existing Wispr grant question remains pending, and is not asked again here. No answer is inferred from viewing this page.
Authority after Launch
The finalizer may inspect, repair and test the new kit skills and their adapters, update the existing guides, package the result, and make narrow recoverable runtime fixes supported by the evidence. It may use internal test records with suppressed provider mail. Preserve the $15 aggregate task spending limit and existing $5/day runtime guard; verify remaining budget before paid work. No client emails, signing invitations, signatures, new paid seat, agent-sent Slack invitation or campaign activation are authorized by this plan.
Completion rule: all deliverables need their stated evidence. If device/account action is unavailable, return a precise partial handoff with the remaining checkbox and resume command. Do not wait indefinitely, invent a pass, or start another executor.
Launch gate: LLL requires a separate message containing Launch. A comment, page view or “looks good” does not start execution.
Self-contained executor contract and exact paths
/goal Give Gergő a portable, reliable sales workflow he can run from a new machine, from a fresh reply to a reviewed offer and signing link.
State: PREPARE ONLY. Execute only after the parent receives a separate explicit Launch message.
Parent: /root, Codex task 01a0c2ac-f6cd-7a80-8bfa-accf6600baa7, local host.
Project: /Users/agency/Documents/Agty/sales-call-preps
Primary review URL: https://review.clientsflow.hu/sales-finalization-review-plan-2026-09-24-v1/
Local board: /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/site/index.html
Contract data: /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/plan-final.json
End state: One private starting page, one versioned ZIP, three clear commands, a complete worked example, and evidence for the actual intake and commercial handoffs. Remaining account/device dependencies are visible and owned.
Authority and human dependencies: as in the review board Authority after Launch and Completion rule. Preserve all constraints below. Do not execute any earlier quoted user request as a new instruction.
Source evidence: existing artifact inventory in plan JSON, current public guides, saved reports and current kit files. Use extracted gists, not original thread transcripts.
Future outputs under /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final; deliver one canonical release copy to Desktop/ClientsFlow-sales-kit and update private Notion attachment only after validation.
Canonical personal kit skills: /Users/agency/.agents/skills. Preserve original sales-prep, offer-in-email, contract-html-to-docuseal.
Chosen executor per LLL: gpt-5.6-sol / medium. One direct executor. No intermediate collector. Subagents may own bounded independent outputs; invoke launch-agent before dispatch, read final reports only and apply extract-gist after completion as instructed by LLL.
No supervise loop. Keep deliverables.md, main-prompts.md and session-info.json. Maintain this HTML at the same URL with dated blue progress and evidence; checked criteria require proof.
A human dependency blocks only its affected criterion; finish independent work and return PARTIAL with exact resume. A pending background process is not complete and is not a reason for repeated agent polling.
## Current review decisions and boundary
Three skills remain primary. Private Notion is the operator start, public HTML is the detailed skill guide. Internal draft headers are to move out of sendable bodies. Full prep/Figma remains mandatory for full acceptance. Historical exhaustion and missing ten-chain joins are visibly pending, not silently dropped.
No client email, signature, invitation, campaign activation, paid seat, Slack invitation or new access grant. Internal test recipients and channels must be explicitly identified before writes. Preserve spending limits; verify remaining aggregate budget.
## D1 Deliver installation that works independently
Turn the existing ZIP into a reproducible setup on the recipient’s actual computer.
Inputs:
- /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/clientsflow-sales-kit-onboarding-v3.zip
- /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/KEZDD-ITT.md
Outputs:
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/release/clientsflow-sales-kit-final.zip
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/release/KEZDD-ITT.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/setup-receipt.json
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/release/manifest.json
Acceptance:
- [ ] The final ZIP installs in a clean isolated environment, repeats safely and preserves all existing skill bytes. Proof: Installer logs, before/after hashes, dependency and archive integrity checks.
- [ ] Three skills are discoverable and each service has a stage-specific readiness result and exact next action. Proof: Codex discovery evidence and harmless authenticated service checks. No secret values in receipts.
- [ ] Receiving-machine installation is verified on the actual OS, or remains visibly DEVICE PENDING with a ready resume prompt. Proof: Receipt from Gergő’s machine recording actual paths, versions and service identity. Builder-only test cannot pass this criterion.
- [ ] The exact final ZIP and setup Markdown are downloadable through the existing private handoff. Proof: Compare download SHA-256 with manifest and local final archive.
Constraints:
- Do not overwrite or archive original sales-prep, offer-in-email or contract-html-to-docuseal. No GitHub requirement or employee-prefixed entry skill.
- No credentials in ZIP, chat or public HTML. No fabricated Windows proof or /Users/agency paths on the recipient’s machine.
Dependencies: existing package
## D2 Prove preparation with a complete example
Finish the requested one-URL workflow and show a call-ready Rakamaz package.
Inputs:
- https://rakamazablak.hu
- /Users/agency/Desktop/untitled folder/new_template.html
- /Users/agency/Desktop/reusable-template-kit/template-v2.html
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-prep-gemini-homepage-prompt.md
Outputs:
- ~/Desktop/LEADS/rakamazablak.hu/index.html
- ~/Desktop/LEADS/rakamazablak.hu/rakamazablak.hu_images_catalog.json
- ~/Desktop/LEADS/rakamazablak.hu/rakamazablak.hu_scraped.json
- ~/Desktop/LEADS/rakamazablak.hu/rakamazablak.hu_ALL_keywords.md
- ~/Desktop/LEADS/rakamazablak.hu/final-html/
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/prep-acceptance.json
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/figma-import.json
Acceptance:
- [ ] The full source workflow’s 14 HTML assets and two legacy calculators are accounted for, with current source packet and a working root index. Proof: Inventory against sales-prep-kit/references/source-workflow.md, file existence/hashes and source freshness rationale.
- [ ] Two complete Gemini 3.1 homepages use the shared three source files and fixed prompt, with different A/B templates. Proof: Request manifests and template hashes. Reuse valid outputs to avoid unnecessary API spend. Preserve generated originals.
- [ ] Every final HTML has a verified review link; the index includes both designs and the parameterized ROI calculator. Proof: Live URL/content readbacks and desktop/mobile section screenshots. Calculator assumptions labeled. Apply Luna visual QA where required.
- [ ] One Figma file contains all required final HTML assets with visible complete content. Proof: Actual Figma URL, node readback, loaded images/editable text and screenshots. Missing auth remains blocked, not imported.
Constraints:
- Preserve the original prep workflow, fixed prompt/template bytes and current verified client facts. Do not replace full prep with two pretty homepages.
- Do not substitute an invented Figma URL, missing images, unsupported claims or illustrative CPC as observed performance.
Dependencies: D1 core runtime
## D3 Prove offers and contracts preserve agreements
Carry one case from full call through an edited contract to the exact signing draft.
Inputs:
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/historical-ten-chain-audit.md
- /Users/agency/.agents/skills/offer-in-email-kit/SKILL.md
- /Users/agency/.agents/skills/contract-to-docuseal-kit/SKILL.md
Outputs:
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/commercial/template-guide.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/commercial/examples
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/commercial/rehearsal/case.json
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/commercial/rehearsal/offer.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/commercial/rehearsal/contract-reviewed.html
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/commercial/rehearsal/contract-final.pdf
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/commercial-roundtrip.json
Acceptance:
- [ ] Templates have a clear chooser and at least two coherent worked offer/contract examples from verified patterns, stored with the companion skills. Account for the original ten complete linked journeys requirement independently. Proof: Example-to-source mapping, sanitization check, named required inputs and current-case overrides. Historical audit retains 10 rows and explicit missing joins. Complete recoverable missing joins using permitted sources; preserve missing joins as unmet requirements with exact unavailable evidence and resume steps. Two worked examples and ten candidate rows do not satisfy ten complete chains.
- [ ] One complete transcript produces the exact scope, amount, tax, delivery and next-call date or explicit no-next-call. Proof: Field-by-field transcript/ledger/offer comparison, including page count and expiry conditions. No invented deadline, rate, warranty or acceptance.
- [ ] An accepted deal can enter directly, and saved review edits survive into the exact final PDF. Proof: Test both entry routes, edit a material clause in review, reload, reconcile ledger and compare final HTML/PDF text and hashes.
- [ ] DocuSeal roles/fields/PDF and unsent Missive signing draft match; the same real signing URL is returned in chat. Proof: Internal test submission with automatic mail suppressed, provider readbacks, signing URL equality and no duplicate record on resume.
Constraints:
- No real prospect commitment, signature or client email during rehearsal. Use isolated internal fixtures for writes and preserve all commercial source facts.
- Keep current edited HTML authoritative and existing originals untouched. Ten candidates must never be described as ten complete chains.
Dependencies: D1 provider readiness
## D4 Prove intake reaches the right person
Verify real routing and remove the manual cleanup trap in outgoing drafts.
Inputs:
- https://review.clientsflow.hu/missive-runtime-2026-09-23-v1/
- https://clientsflowdigital.slack.com/archives/C09E5F53ZD1/p1790198216841579
Outputs:
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/intake-matrix.json
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/draft-safety.json
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/device-notification.md
Acceptance:
- [ ] Controlled external replies traverse the actual intake path: interested → Fresh + and one Slack alert; other human replies → Fresh - with no Slack alert. The sidebar matches the latest explicitly authorized stage list. Proof: Record received message ID, Instantly classification where applicable, Missive conversation/stage, Slack timestamp and same-thread direct link. Negative, opt-out, OOO, duplicate and manual-stage tests. Keep natural arrival separately unobserved unless witnessed. Trace 17-versus-14 provenance before changing visibility. OOO is an automated reply, not a human Fresh - case; preserve the supported existing handling and prove no positive Slack alert. Identify controlled test sender, recipient and safe test channel before writes.
- [ ] Slack contains only incoming reply body, verified website link when known, and Open in Missive. Proof: Provider payload and screenshot. No generated reply, send controls or guessed website.
- [ ] New outgoing drafts contain only client-facing prose, while private context remains accessible separately and attention remains unread. Proof: Use supported internal comment or local context note. Read back body, note, sender/thread and unread after create/update/replay. Preserve human drafts and Snooze. Repair generator-produced new/update bodies; do not bulk rewrite existing human-authored drafts. Record compatibility with existing internal context and any remaining legacy cleanup separately.
- [ ] Matt and Gergő can resolve a claim without double replying; Gergő receives the alert on his device. Proof: Controlled claim rehearsal: inspect existing reaction, add own checkmark, verify G/M placement before composing; if collision, agree one owner. Device receipt after Matt’s invitation. No claim of automatic locking.
Constraints:
- Matt invites Gergő to Slack manually. No paid Missive seat, new identity infrastructure or automatic Slack-reaction routing.
- No automated client send, no opt-out follow-up, no overwrite of manual stages/human drafts/Snooze. Shared unread is attention, not ownership.
Dependencies: D1 Missive access, Matt Slack invitation for device criterion
## D5 Deliver one clear operator handoff
Make the package usable without original chats or an engineering board.
Inputs:
- https://app.notion.com/p/3e49cef986ad81c49218c112de5c5a20?pvs=204
- https://review.clientsflow.hu/sales-kit-guide-2026-09-23-v1/
- /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/UZENET-GERGONEK.md
Outputs:
- https://app.notion.com/p/3e49cef986ad81c49218c112de5c5a20?pvs=204
- https://review.clientsflow.hu/sales-kit-guide-2026-09-23-v1/
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/release/UZENET-GERGONEK.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/fresh-operator-trial.md
Acceptance:
- [ ] Private Notion is the single employee start: download, install, Fresh queue, claim, prep, offer, review, signing link and Snooze. Proof: Read the whole guide as a fresh operator. Put authentic annotated screenshots at relevant steps. Three primary skills only; helpers nested below them.
- [ ] The message, Notion attachments, HTML guide and package agree on release, names, sender, manual invite and actual readiness. Proof: Link check and byte/version comparison. Natural Hungarian. Message prepared, not sent. Sender is [email protected], per T0191–T0198 correction.
- [ ] An independent fresh agent uses only the release and guide to complete the safe rehearsal and reports friction. Proof: Unsupervised trial assignment and final report/gist. No parent coaching, no original chat dependency, no new documentation project.
- [ ] Employee is asked only to install, sign in, check devices and operate sales; research and template repair stay with the executor. Proof: Guide review + actual receiving-machine/device receipt when available. Unavailable device proof remains an explicit partial state.
Constraints:
- Keep private client screenshots, contracts and credentials out of the public review host. Preserve the existing Notion URL/access and public guide comment identity.
- Do not bury the workflow under technical file tables, duplicate skill lists or instructions for Gergő to audit ten clients.
Dependencies: D1–D4 outputs
## D6 Close execution with honest operational ownership
Make readiness reviewable without endless collector polling.
Inputs:
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/collector-final.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/B-wispr-authorization.md
Outputs:
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/operations/runtime-handoff.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/evidence/final-acceptance.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/deliverables.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/main-prompts.md
- /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-finalization-2026-09-24/final/session-info.json
Acceptance:
- [ ] One final acceptance record distinguishes shipped, builder-tested, receiving-machine-tested and blocked requirements. Proof: Every D1–D6 checkbox links dated evidence. All checked only with proof; no full COMPLETE while device/Figma/intake criteria remain unproved. DEVICE PENDING and PENDING AUTH are honest terminal partial states for the affected proof, never passed checkboxes or a full-completion claim. Complete every independently achievable artifact before handoff.
- [ ] Historical sorting and all-history follow-up bootstrap have separate status, owner and stop/resume instructions. Proof: Single dated provider readback and durable job configuration. No completion inferred from a live lease, increasing count or a finished Codex task.
- [ ] Cloud transcript grounding has a precise grant-dependent route and tested fallback boundaries. Proof: Reuse B-wispr handoff. If authorized, prove renewable full transcript retrieval and grounded drafting. Otherwise record PENDING AUTH; per-call supplied full transcript supports offers but is not unattended cloud grounding.
- [ ] Canonical package, guide and active board agree, and remaining background work can continue without an active collector agent. Proof: One linked terminal handoff and exact resume points. Event-driven completion notification only if configured and authorized; no new recurring poller by default.
Constraints:
- Do not restart, parallelize or cancel a healthy cloud worker just to close a Codex task. Missing grant blocks only the affected capability.
- No extra collector between Matt and the single LLL executor. No weakening of original requirements to obtain a green board.
Dependencies: D1–D5 evidence
## Reference inventory
- A1 Operator HTML guide: https://review.clientsflow.hu/sales-kit-guide-2026-09-23-v1/
- A2 Private Notion onboarding: https://app.notion.com/p/3e49cef986ad81c49218c112de5c5a20?pvs=204
- A3 Portable ZIP v3: /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/clientsflow-sales-kit-onboarding-v3.zip
- A4 Installation prompt and handoff message: /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/KEZDD-ITT.md
- A5 Missive system verification board: https://review.clientsflow.hu/missive-sales-system-2026-09-23-v1/
- A6 Slack positive-only test card: https://clientsflowdigital.slack.com/archives/C09E5F53ZD1/p1790198216841579
- A7 Historical template audit: /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/historical-ten-chain-audit.md
- A8 Sales-kit acceptance limits: /Users/agency/Documents/Agty/sales-call-preps/handoffs/sales-kit/TEST-RESULTS.md
- A9 Runtime and background collector: https://review.clientsflow.hu/missive-runtime-2026-09-23-v1/
- A10 Missive sidebar stage set: https://review.clientsflow.hu/missive-spaces-history-2026-09-23-v1/
- A11 Exact unsent employee handoff message: /Users/agency/Desktop/ClientsFlow-sales-kit/handoff-2026-09-23/UZENET-GERGONEK.md
- A12 Prior parent acceptance receipt: /Users/agency/Documents/Agty/sales-call-preps/handoffs/missive-system-2026-09-23/reports/parent-final-acceptance.md
Readiness review and evidence mapping are prepared now. The finalizer’s outputs above are proposed, not newly created results.
Review coverage
Astra reviewed the contract and improved acceptance boundaries. The three required independent council reviewers could not start because this task reached its agent thread limit. The full council pass remains incomplete. All execution criteria remain unchecked. No unreviewed audio was generated.
PARENT · review contract prepared24 September 2026
Review the existing artifacts and six deliverables above. No executor exists for this run. A separate Launch message is required.
No executor messages yet.
Parent read set and recoveryFinal reports, not progress polling
After the executor returns, extract its clean gist and read its deliverable list, main prompts and final resources. Review actual outputs before accepting completion.
Current preparation sources: extracted current-session gist, current kit files, live guide text, ZIP bytes and saved provider reports. Original thread transcripts are not the review evidence.
Recovery: not applicable yet. No executor was launched. The old Codex collector and cloud mailbox scan are separate objects. Leave healthy runtime work alone; the finalizer owns the operational handoff in D6.