Know-How OSmock v0.1
{{ stepLabel }}
{{ stepTitle }} {{ stepDesc }}
Know-How OS
invite only · pilot
ON-CALL HUMAN EXPERTS · FOR AI AGENTS

Your agent hits a wall.
A verified human picks it up in minutes.

Know-How OS is an MCP server and a REST API. The agent sends a fixed-price contract. A checked expert accepts inside the SLA and returns a structured result. No browsing, no bidding, no chat with a bot.

mcp · knowhow.os/mcp · tool call
→ request_expert({
  task_type: "code_review",
  goal: "Is PR #412 safe to merge? Focus on token reuse.",
  inputs: ["github.com/lumen-labs/api/pull/412"],
  price: 12000,
  sla: { accept_within: 30, deliver_within: 240 }
})
← { job_id: "job_8f2a", status: "matched",
    expert: "exp_7f3a", tier: "T3",
    accepted_after: "6m 12s", deliver_by: "13:42 UTC" }
01 · CONTRACT
The agent fills a strict schema
Goal, inputs, deliverable, acceptance checks, price, SLA. If it is not in the contract, the expert may decline.
02 · MATCH
The platform picks the person
Field, tier, time zone and NDA status decide who is notified. The first qualified expert to accept gets the job.
03 · DELIVER
A result the agent can parse
Machine checks run first, then the agent accepts or asks for one revision. Paid from a prepaid balance on acceptance.
14 min
median time to expert acceptance
2h 40m
median time to delivery
87%
accepted without revision
1.2%
disputes · pilot, last 30 days

Twelve fields. Each with its own deliverable schema.

Every field is a template: what the agent sends, what the expert returns, the machine checks, the price band, the default SLA.

{{ f.name }}{{ f.n }} on call
{{ f.key }} · {{ f.band }}
{{ f.note }}

Every expert is a person we checked.

Verification is done by people in the pilot. The software records what was checked, by whom, and when.

{{ c.n }}
{{ c.t }}
{{ c.d }}
WHAT THE AGENT SEES
7f
exp_7f3a
T3 · Principal · trust 94/100
Verified fields
Code & security · Compliance
History
62 jobs · 4.9 · 0 disputes lost
Identity
ID + video call · 12 Sep 2026
Name, contact
held by the platform
For experts: a job card, a timer, a form, a payout.
No bidding, no proposals, no chat. You join by referral from a live expert.
Know-How OS · pilot · USD only · compliance, legal and clinical fields are review, not advice.
EXPERT VERIFICATION
Invited by Dana K. T3
Usually 3 to 5 days end to end. The video call and the trial job are the slow parts.
STEP 1 OF 7

Dana K. invited you as a code and security reviewer.

Know-How OS is invite only. Dana vouches for you; if you pass verification, your first ten jobs count toward her referral record too. Here is what the next steps involve and why.

Identity
ID document and a 10-minute video call. We store that it passed, not the document.
Proof per field
We check your claims against the source: GitHub, references, samples.
Skill test
A real, anonymised past job with seeded issues. Graded blind by two T3 experts.
Paid trial job
One live job, graded by the operator. Then you are on call.
Invite code KH-DK-0412 · expires in 6 days
STEP 2 OF 7 · IDENTITY

Prove you are a real person.

Agents never see your name. Buyers never see it either. The operator sees it once, on this call, and the platform holds it for contracts and payouts.

government id
Passport or national ID. Deleted after the call.
10-minute video call
With the operator. We match your face to the ID and talk through your proof.
storedthat the check passed, the operator's name, the datenot storedthe document, the recording, your face
STEP 3 OF 7 · EXPERTISE

Define your own territory.

Pick the fields you want jobs in, then narrow each one. Matching uses the narrow skills, not the broad field, so be specific. Every claim here is something we will check in the next step.

Fields · {{ fieldCount }} selected
Specific skills in {{ activeFieldName }}
You can add fields later; each one goes through proof and test again.
STEP 4 OF 7 · PROOF

We check claims against the source, not the CV.

Connect the places where your work already lives. We read signals, compare them to what you told us in the last step, and flag anything that does not line up for the video call.

GitHub · for Code & security review
Read-only. We look at review activity, not stars.
connected · @nhaddad
11 y
account age
1,240
PR reviews authored
38%
of reviews in auth / payments paths
4
CVE credits
References · for Compliance & policy
We email them a 3-question form. Work addresses only.
M. Okafor · Head of Risk, Fenwick Payreplied ✓
L. Brandt · GC, Orbit Healthsent · 2d
Cross-check
{{ x.claim }}
{{ x.source }}
{{ x.status }}
One open item goes to the video call. Nothing here blocks you from the test.
STEP 5 OF 7 · SKILL TEST

A real job, anonymised, with issues we planted.

You get the same job card an agent would send. Two T3 experts grade it blind. Nobody can study for it because every test is a different past job.

test · code_review · medium
Review a 640-line diff adding SSO to a billing service
45 minutes. Same deliverable form as a live job: verdict, findings with severity and location, one-line fix each, summary.
5 seeded issues·pass ≥ 75·paid $0
Passed86/100
Seeded issues found4 / 5
False positives0
Fix quality (graders)9 · 8 / 10
Time used38 min
Missed: the IdP metadata cache never expires (medium). Both graders noted clear, actionable fixes.
The timer starts when you open the diff. You can take it any time in the next 7 days.
STEP 6 OF 7 · TRIAL JOB

One live job, paid, graded by the operator.

A real buyer's agent, a real SLA, real money. The operator reads your result before the buyer does. This is also your first look at the expert app on your phone.

job_7c0e · code_review · small · trial
Check a webhook signature verifier before it ships
accepted · paid $48
Accepted after
9 min
Delivered after
1 h 52
Revisions
0
Operator grade
A−
Operator note. Found the timing-unsafe compare and the missing replay window. Summary could be one line shorter. Clear to go live.
STEP 7 OF 7 · GO LIVE

You are exp_3k9d. Decide how you want to work.

Agents see the handle, the tier, the trust score and the verified fields. Everything below is yours to change at any time.

Trust score71/100
{{ t.k }}{{ t.pts }}
{{ t.v }}
T1 now · T2 after 10 accepted jobs, rating ≥ 4.6, no lost disputes · T3 after 50 jobs and grading others' tests
Work you accept
Notify me on
Email is always the fallback. Payouts weekly via Wise, min $50.
Legal identity: Noor Haddad · held by Know-How OS, shown to no buyer or agent.
EXPERT APP · exp_7f3a · T3
Know-How OSon call
On call
until 18:00 · Code & security, Compliance
No open jobs. We will ping you on Telegram.
THIS WEEK
4 jobs accepted$412
Median accept5 min
Payout Friday$1,096
New job · code_review · small · $96 net
job_8f2a · Lumen Labsaccept in 23:48
Confirm the refresh-token rotation in PR #412 is safe to merge. Focus on reuse detection and session revocation.
inputsPR #412 · diff 142 lines, 3 files · pnpm test --filter auth
returnverdict · findings (severity, file:line, fix) · summary
deliverwithin 4 h of accepting
ndacovered by your platform NDA
You receive$96.00 of $120 · 20% fee
job_8f2a · in_progressdeliver in 3:52:10
INPUTS
github.com/lumen-labs/api/pull/412Open
pr-412.diff · 142 linesOpen
pnpm test --filter authruns on submit
DELIVERABLE
Verdict
Approve
Changes
Reject
Findings · 0
Summary
One paragraph the agent can read as-is…
job_8f2a · delivered2 h 41 after accept
Verdictchanges_needed
{{ f.sev }}{{ f.loc }}
{{ f.title }}
fix {{ f.fix }}
Summary
Rotation works, but a stolen old token still refreshes once. One fix, about 10 lines, then safe to merge.
ACCEPTANCE CHECKS
All findings have severity + locationpass
Summary presentpass
pnpm test --filter authran · 48/48
Agent or buyer acceptswaiting · auto in 23:49 h
job_8f2a · accepted · paid
+$96.00
Accepted by the agent 11 min after delivery · no revision
LEDGER
job_8f2a · code_review
Lumen Labs · today 12:03
+96.00
job_7e11 · compliance
Orbit Health · yesterday
+64.00
job_7d9c · code_review
Nimbus · Mon
+160.00
Payout Friday · Wise$1,192.00
Rated 5 by the buyer · trust 94 → 94
On call
until 18:00 · Code & security, Compliance
ENGAGEMENTS
None yet. You accept engagements in your settings; buyers' agents can propose one when your profile fits.
RECENT
job_8f2a · code_review
Lumen Labs · paid
+96.00
job_7e11 · compliance
Orbit Health · paid
+64.00
Engagement offer · Lumen Labs · 12 weeks
eng_2c1d · weekly · code_reviewrespond in 23 h
Weekly security review of everything merged to main, 12 weeks from 2 Nov.
scopeAll PRs merged that week, up to 1,000 changed lines. Larger weeks split into two jobs.
windowInputs Mon 09:00 UTC · deliver by Tue 17:00 UTC
why youReviewed PR #412 for this buyer · auth, payments · T3
docsMutual NDA + engagement contract, drafted by the platform
Per review, net
$144
12 weeks, net
$1,728
eng_2c1d · documents
{{ c.n }}{{ c.t }}{{ c.tag }}
{{ c.text }}
Platform standard is 6 months.
Changed to 6 months · accepted by the buyer's agent 4 min ago (within buyer policy)
One change request per document. Signing is next.
eng_2c1d · sign
You are signing two documents.
Mutual NDA · v3 · 7.2 amendedread ✓
Engagement contract · v2read ✓
You sign as exp_7f3a. Your legal name is on the platform's copy only; Lumen Labs contracts with Know-How OS, which contracts with you. Identity verified 12 Sep 2026.
Type your name to sign
Buyer side: signed by J. Park, Lumen Labs (a person, not the agent) · 09:14 UTC
Lumen Labs · weekly security review
eng_2c1d · active · week 1 of 12
nextInputs arrive Mon 2 Nov 09:00 UTC · job_9a01 already created, $180 held
deliverby Tue 3 Nov 17:00 UTC
docsNDA · Contract · signed 26 Oct
Committed, net$1,728
Paid so far$0
End early1 week notice, either side
Your on-call hours are blocked Mon 09:00–Tue 17:00 for this engagement.
AGENT · lumen-labs/deploy-bot · via MCPbalance $4,180 · limit $300/job
agent
{{ m.text }}
→ tool call · {{ m.name }}
{{ m.body }}
← {{ m.name }}
{{ m.body }}
platform{{ m.text }}
← search_experts · 4 candidates · sorted by trust score · names withheld
{{ c.ini }}
{{ c.h }}{{ c.tier }}{{ c.fields }}
{{ c.hist }} · {{ c.tz }} · {{ c.avail }} · {{ c.rate }}
{{ c.note }}
{{ c.trust }}
trust
{{ waitingText }}
Press → or Play to advance. The agent's text is a simulation; the tool calls and payloads match the contract spec.
MOCK NOTES · READ WITH THE PRD v0.1

What this mock assumes, and where it leaves the PRD.

Kept from the PRD

  • Agent hires human through MCP and REST; the contract schema is the product.
  • Platform matches. No browsing, bidding, proposals or chat for short jobs.
  • Fixed price, prepaid balance, 20% take rate, auto-accept after 24 h, one revision round.
  • Invite only on both sides. Expert identity held by the platform, handle shown to agents.

Added for the vision discussion

  • Twelve fields instead of two. Each is still a template with its own deliverable schema; legal and clinical are labelled review, not advice.
  • Engagements. A long-term, cadence-based contract. The agent uses search_experts and propose_engagement; the platform drafts the NDA and the contract from templates, the expert gets one change request per document, both sides e-sign. Each week then becomes a normal job.
  • Three-party contracting. Buyer contracts with Know-How OS, Know-How OS contracts with the expert. That is what lets the expert stay anonymous while still signing something enforceable. Needs a lawyer.
  • Tiers T1 / T2 / T3 and a trust score with four parts: identity, proof, skill test, history. The score is explained, never a black box.
  • Anti-fake onboarding. Claims are typed first, then checked against sources (GitHub activity, references, platform-run test with seeded issues, graded trial job). Mismatches go to the video call rather than silently failing people.
  • Expert-defined territory. Fields, narrow skills, a one-line edge, working preferences (job kinds, min price, concurrency, hours, channel). Matching uses the narrow skills.

Mocked, not designed

  • Buyer console and operator admin (not in this round).
  • Search and filter UI for human buyers; only the MCP path is shown.
  • Decline, revision and dispute paths on the phone.
  • Real timers; every countdown is a still frame.

Questions this raised

  • Who may propose an engagement: any agent, or only after one accepted short job with that expert?
  • Should experts see the buyer's name on engagements, given they sign an NDA with them?
  • Does a broad field list dilute the pilot's "two task types" test, or is it only the landing page's promise?
  • What does the operator do when the skill test and the GitHub signals disagree?