Command
Script output
← Operation Vela
Open source plugin for Claude Code · v1.1.0
Malha GTM
Personas, positioning and messaging that don't make up evidence.
You talk to customers. Malha GTM turns what they said into sourced facts and delivers everything connected: who buys, what they want, what you offer and how to say it. Any claim traces back, through a trail, to a customer's sentence.
- 11skills: 1 master and 10 stage skills
- 21automated gap checks
- 87tests passing
- 0made-up quotes: a script checks every one
In 6 frames
It all starts with a conversation
Collection brings together calls, team answers, tickets and the CRM.
Call · call-2026-08-04-rota-leste.txt
Vela: What made you start looking for a system now?
“I close payroll for 140 people across three spreadsheets and send everything to the accountant to check.”
F-0001 · evidence 2Payroll for 140 people closed across three spreadsheets00:01:24
F-0002A broken formula paid the wrong overtime
F-0016Overtime discovered only after the payslip
Finding A-02Unapproved overtime is the surprise cost that hurts the mostconfirmed
PersonaPayroll analyst who closes payroll and answers for any error
Value theme · with proofPayroll closed with no surprises: the mismatch shows up before payday
Messaging pillar“Close payroll only once, with no surprises on payday”
End-to-end test
Malha GTM running on Vela RH, from an empty folder to messaging
The plugin's scripts really ran, step by step. The company and the answers are fictitious, but every quote matches the files letter by letter: the validator checked. The plugin works in Portuguese: commands and script outputs are shown as recorded, and the quotes are translated.
- 14sources
- 61facts with quotes
- 9findings
- 233nodes in the graph
- 276links
- 2times a gate blocked
What came out of the test
Every field points to a customer's sentence
Persona
Payroll analyst who closes payroll and answers for any error
SMBs with 50 to 300 people · 1 to 2 people in payroll · reports to the owner or the finance director
- Why now
- Stop redoing payroll because of spreadsheet errors and overtime discovered too late
- What success looks like
- Closing without redoing it after payday
- What holds them back
- An accountant who may not accept the file, the spreadsheet habit, app adoption
- What decides it
- Seeing the mismatch before closing; time entries flowing into payroll on their own
- How they buy
- Tests it with one department for a week and takes it to the owner
Positioning
The core difference: Vela only lets you close when time tracking, benefits and payroll match.
- Spreadsheet plus the accountantFamiliar and no extra monthly fee; rework, formulas that break and overtime discovered too late.
- FolhexA suite with recruiting and performance; too big for SMBs, charges per module and takes months to roll out.
- Legacy desktop payroll systemAlready calculates payroll; no mobile time tracking and no check before closing.
- Hiring someone to check time entriesSolves it with people; a fixed cost for a repetitive problem.
The gate that blocked the website's promise
“Rollout in 2 weeks, with support from real people”
It was on Vela's website, but no customer confirmed it, and two quotes said the opposite. Malha GTM didn't let it become a message: the theme was removed and became a hypothesis.
“The first payroll closed in three weeks because the accountant took a while to validate the file”tickets-2026-t3-ficticio.csv · row 5
“My worry is the rollout getting in the way of the September close.”call-2026-08-19-bom-ar.txt · 00:07:30
The path
You only call the Malha GTM master skill. It runs the rest.
The master reads the project state, decides which step comes next and only lets you move forward when the gate is open. It picks up where it left off, even in another session.
- 1 Collection
Generates a questionnaire per team, logs the attachments and scores an 11-criteria rubric. Asks at most 3 things per round.
Gate: None: it's the way in.
- 2 Evidence breakdown
Breaks each source into facts with a literal quote, location and tags. Groups them into findings; confirmed or hypothesis is calculated by triangulation.
Gate: Green rubric (items 1 to 5 and 9) or collection closed.
- 3 ICP and buying committee
Account ICP, negative ICP, triggers and one card per buying committee role.
Gate: Evidence breakdown complete.
- 4 JTBD
The company's job and each role's job, the four forces with quotes and the job map.
Gate: ICP and committee complete.
- 5 Personas
Role + circumstance + jobs, with the 5 rings of buying insight, pains, gains and objections, each field with its own confidence.
Gate: JTBD complete.
- 6 Journey
Eight stages per persona crossed with the buying jobs and touchpoints.
Gate: Personas complete.
- 7 Value proposition
One canvas per persona and job, with each pain reliever linked to the pain it solves.
Gate: Journey complete.
- 8 Positioning
In April Dunford's order: alternatives, unique attributes, value themes with proof and category.
Gate: Value proposition complete, or the shortcut right after the evidence breakdown.
- 9 Messaging
Message house, persona × stage matrix, pitch and assets that answer objections.
Gate: No value theme without proof and no persona without a status quo.
The evidence
What the customer does is worth more than what they say
A finding only becomes confirmed with 2 types of customer source or 3 different customers. Internal opinion never confirms anything.
- 0 · Assumptioninternal opinion with no example
- 1 · Internal anecdotesomeone at the company cites a case
- 2 · Customer saidan interview or review quote
- 3 · Customer didCRM, ticket, usage, churn
- 4 · Customer paidclosed against an alternative
21 checks: the graph as auditor
- blockingFact without a source, or a quote that doesn't exist in the file
- blockingPersona without a status quo alternative
- blockingValue theme without a capability or without proof
- blockingICP that can't be turned into a list
- blockingEvidence-0 node without the HYPOTHESIS stamp
- blockingGraph outside the schema
- warningMaximum-severity pain without a pain reliever
- warningOrphan feature: pain reliever without a pain
- warningOrphan capability
- warningValue map that addresses 100% of the profile
- warningProof based on a finding with contrary evidence
- warningMore than 3 value themes
- warningRole with veto power and no resolved objection
- warningChampion with an empty cell in the matrix
- warningInsight identical to a generic trend
- warningTrend without a category
- warningNearly identical rows in the messaging matrix
- warningPersona without the 5 rings or job without the 4 forces
- warningPurely functional profile, or a gain that mirrors a pain
- noticeWhy–how ladder outside 2 to 4 rungs
- noticeExperience incompatible with the cost of error
How it was built
Four phases, each approved before moving on
- Phase 1 · MVPMaster, collection, evidence breakdown, positioning and graph, with an offline viewerWhat the test revealed: The master stopped at the end of the sub-skill: the close command was born, which gives the next order.
- Phase 2ICP and committee, JTBD, personas and the Dunford shortcutWhat the test revealed: The model cherry-picked proof: the contested-proof check was born.
- Phase 3Journey, value proposition, messaging and the messaging gateWhat the test revealed: The model justified a warning: the master gained the rule “a warning can't be justified away”.
- Phase 4README in Portuguese and English, contributing guide, install catalogWhat the test revealed: Report with no local paths; version 1.0.0.
Install
/plugin marketplace add <usuario>/malha-gtm /plugin install malha-gtm@malha-gtm
The repository goes live soon.
Known limits
- Validation guarantees the quote exists, not that it was read correctly.
- The graph schema and the rubric thresholds still need real-world use.
- The model can cherry-pick evidence; human review is still necessary.