TMR Land , Personal Skill

connect your agent to tmr land, a bilingual (zh/en) ai business marketplace. as a personal user you search businesses, publish intentions, place escrow orders, and evaluate business quality via credit scoring. the platform uses multi-path matching (rules, bm25, vector, rrf fusion) to surface candidate businesses, negotiation-based deal flow, and an escrow payment system with delivery confirmation and dispute resolution.

intent

use this skill to operate as a personal buyer on tmr land's ai business marketplace. it covers five core workflows: (1) searching for ai and data service businesses without committing to an intention, (2) publishing structured purchase intentions and triggering multi-path matching, (3) negotiating deals with candidate businesses via message-based sessions, (4) placing and managing escrow orders from intention through payment, delivery, and review, and (5) evaluating business quality via credit profiles and reputation scores. use the skill whenever you need to source, negotiate, or transact on ai/data services.

inputs

required environment variable

• TMR_API_KEY (string, env var): personal-role api key. create via POST /api/v1/api-keys with role: "personal". bearer token auth for all requests.

optional environment variable

• TMR_BASE_URL (string, env var, default: https://tmrland.com/api/v1): api endpoint base url. override for staging or custom deployments.

runtime inputs (user-provided)

• intention content (string): natural language description of what you need (e.g., "fine-tuned chinese nlp model for sentiment analysis")
• locale (string, auto-detected): zh or en, inferred from intention content language
• business ids (array of uuids): selected from match results for negotiation
• order ids, session ids, intention ids (uuids): workflow identifiers
• wallet currency (string): USD or USDC for payment and wallet operations
• message text (string): user-authored negotiation or order chat
• review rating (integer 1-5): user's assessment of business performance
• kyc details (object): name, id-type, id-number for identity verification

external connections

• tmr land api v1: endpoint for all marketplace operations. requires valid personal-role api key. rate limits: 100 requests per minute (check rate-limit headers on responses). auth tokens expire after 24 hours; refresh by creating a new key if needed.

contextual data

• current intention state (draft/published/matching/matched/negotiating/contracted/cancelled/expired)
• current order state (pending_payment/delivering/pending_review/revision_requested/pending_rating/completed/disputed/cancelled/refunded)
• current negotiation state (active/contracted/rejected/cancelled/closed)
• wallet balance (usd and usdc)
• kyc verification status (pending/approved/rejected)

procedure

core workflows

workflow 1: quick business search (no intention)

1. input: user provides search content (e.g., "need a sentiment analysis model")
2. call: quick-search.mjs --content "<content>" (synchronous, returns matches directly)
3. output: array of matched business profiles with reputation and credit summary
4. display: show user match score, business name, reputation, review count
5. next: user can view business detail or start intention-based workflow

workflow 2: intention-based matching (structured need)

1. input: user describes need via natural language
2. agent action: draft intention content (confirm with user before submitting)
3. call: create-intention.mjs --content "<content>" [--locale zh|en]
4. input/output: create response contains intention_id (uuid)
5. call: publish-intention.mjs <intention_id> (moves intention from draft to published)
6. output: intention now ready for matching
7. call: trigger-match.mjs <intention_id> (async, returns immediately with match job id)
8. polling loop: call match-status.mjs <intention_id> every 2-3 seconds until status is 'completed' or 'failed'
9. if completed: call get-matches.mjs <intention_id> to retrieve candidate list (sorted by match score)
10. output: array of matched businesses with score, reputation, credit vector
11. display: present top 5-10 candidates to user, including business name, score, review count, credit tier

workflow 3: negotiation flow

1. input: user selects 1-3 businesses from match candidates
2. agent action: collect business ids from user selection
3. call: start-negotiation.mjs --intention <intention_id> --businesses <id1,id2,...>
4. output: array of new negotiation session objects (one per business), each with session_id
5. agent loop: for each session:
   • call: negotiation-messages.mjs <session_id> to check for business proposals
   • display: show latest message and proposal (if any) to user
   • if user wants to message: agent drafts response, shows to user for confirmation, calls negotiation-messages.mjs <session_id> --send "<message>"
   • if business sends final_deal proposal: display terms (amount, currency, delivery timeline), ask user to accept or reject
6. input: user selects one business to close deal with
7. agent action: confirm: "accepting creates an order for [amount] [currency] with [business_name] and cancels all other negotiations. proceed?"
8. call: accept-deal.mjs <session_id> (creates contract and order, cancels other active negotiations for this intention)
9. output: order object with order_id, status 'pending_payment'
10. next: proceed to payment

workflow 4: order payment and delivery

1. input: user is at 'pending_payment' status on an order
2. agent action: show order details (amount, currency, business name, deliverables summary)
3. agent action: confirm: "this will debit [amount] [currency] from your wallet. pay with USD or USDC?"
4. input: user selects currency
5. call: pay-order.mjs <order_id> --currency <USD|USDC>
6. output: order transitions to 'delivering'
7. polling: agent periodically calls order-status.mjs <order_id> to check delivery status
8. when business marks delivery: order transitions to 'pending_review'
9. agent notifies user and displays deliverable summary
10. agent action: ask user to accept delivery or request revision
11. input: user selects accept or revision
12. if accept: agent confirms: "are you sure you want to accept delivery and release [amount] [currency] to [business_name]? this cannot be undone except via dispute."
13. call: accept-delivery.mjs <order_id> (releases escrow, moves to 'pending_rating')
14. output: order transitions to 'pending_rating'
15. if revision: call request-revision.mjs <order_id> --feedback "<user_feedback>" (order returns to business, status 'revision_requested')

workflow 5: review and rating

1. input: order is in 'pending_rating' state
2. agent action: prompt user for rating (1-5 stars) - do not generate or guess
3. input: user provides rating number
4. agent action: suggest comment text based on delivery experience, show to user for confirmation
5. call: submit-review.mjs --order <order_id> --rating <1-5> --comment "<user_comment>"
6. output: order transitions to 'completed'

workflow 6: business research and credit evaluation

1. input: user wants details on a specific business (from match results or search)
2. call: get-business.mjs <business_id> (returns full profile: name, description, services, contact)
3. call: get-reputation.mjs <business_id> (returns reputation scores: review_avg, dispute_rate, completion_rate)
4. call: get-credit.mjs <business_id> (returns credit summary: score, tier, review_count, dispute_count)
5. call: get-credit-profile.mjs <business_id> (returns vector data for agent analysis: review_sentiment, payment_reliability, delivery_speed, communication_quality)
6. call: get-reviews.mjs <business_id> (returns paginated reviews: rating, comment, reviewer, date)
7. output: consolidated business profile with all signals
8. display: show credit tier, review average, key metrics, sample recent reviews

workflow 7: grand apparatus (community Q&A and predictions)

1. input: user wants to see predictions or community answers on a topic
2. call: list-questions.mjs [--limit 10] (returns paginated questions with answer counts)
3. display: show questions, user votes, top answers
4. call: vote-answer.mjs <answer_id> --direction like|dislike (if user wants to signal helpful answer)
5. output: vote recorded

workflow 8: wallet and account management

1. input: user wants to check balance or fund account
2. call: get-wallet.mjs (returns balance in USD and USDC)
3. display: show available balance and currency breakdown
4. if funding needed: call charge-wallet.mjs --amount <N> --currency <USD|USDC> (external payment processor interaction)
5. output: wallet transaction created, funds added after processor confirmation
6. call: list-transactions.mjs [--limit N] to show transaction history
7. input: if user needs kyc: call submit-kyc.mjs --name "..." --id-type <passport|license|id> --id-number "..."
8. output: kyc submission created, status 'pending'
9. call: get-kyc.mjs to check status

management operations

intention lifecycle

• view drafts: list-intentions.mjs [--limit N]
• view details: get-intention.mjs <intention_id>
• update draft: update-intention.mjs <intention_id> [--title "..."] [--description "..."] (only in draft state)
• redescribe (triggers re-match): agent warns user "this will cancel all current negotiations and start fresh. all negotiation progress will be lost. continue?", then calls redescribe-intention.mjs <intention_id> --content "..." [--locale zh]
• delete intention: agent asks "this will permanently delete this intention. this cannot be undone. confirm?", then calls delete-intention.mjs <intention_id>
• cancel intention: agent asks "cancel this intention?", then calls cancel-intention.mjs <intention_id>

order management

• list orders: list-orders.mjs [--limit N]
• check status: order-status.mjs <order_id>
• cancel order (before payment only): agent asks "cancel this order?", then calls cancel-order.mjs <order_id>
• messages: get-messages.mjs <order_id> to view, send-message.mjs <order_id> --content "<text>" to send (agent shows draft to user first)
• mark messages read: mark-messages-read.mjs <order_id>
• get receipt: get-receipt.mjs <order_id>

negotiation management

• list sessions: list-negotiations.mjs [--intention <id>]
• view details: get-negotiation.mjs <session_id>
• view messages: negotiation-messages.mjs <session_id>
• send message: negotiation-messages.mjs <session_id> --send "<text>" (agent shows draft to user first)
• mark read: mark-negotiation-read.mjs <session_id>
• reject proposal: agent asks "reject this proposal? the business can send a revised offer.", then calls reject-deal.mjs <session_id>
• withdraw proposal (if user sent one): withdraw-proposal.mjs <session_id>
• cancel session: agent asks "cancel negotiation with [business]?", then calls cancel-negotiation.mjs <session_id>

dispute and revision

• request revision: request-revision.mjs <order_id> --feedback "<feedback>"
• open dispute: create-dispute.mjs <order_id> --reason "..." [--refund-type full|partial] [--refund-amount N]
• list disputes: list-disputes.mjs [--limit N]

notification and state tracking

• list notifications: list-notifications.mjs
• mark read: mark-notification-read.mjs <notification_id>
• mark all read: mark-all-read.mjs
• unread count: unread-count.mjs

contract tracking

• list contracts: list-contracts.mjs [--limit N]
• view contract: get-contract.mjs <contract_id>

decision points

api key validation if TMR_API_KEY is missing or invalid, all operations fail with 401. agent must ask user to set env var or provide key. if key lacks personal role, operations fail with 403.

locale detection if user provides content in chinese characters, auto-set locale: zh. if english, locale: en. if mixed or ambiguous, ask user.

match status polling when trigger-match.mjs is called, matching is async. agent must poll match-status.mjs until status is 'completed' or 'failed'. if 'failed', display error and offer to retry or redescribe intention. timeout after 60 seconds of polling (approximately 20-30 poll cycles) and inform user that matching is taking longer than expected.

empty match results if get-matches.mjs returns empty array, inform user that no businesses matched the intention. offer to redescribe intention with different keywords or broaden criteria.

wallet balance check before calling pay-order.mjs, agent should call get-wallet.mjs and verify sufficient balance in requested currency. if insufficient, display balance and ask user to charge wallet first.

kyc requirement if user attempts to pay or withdraw and kyc status is not 'approved', call get-kyc.mjs and inform user that kyc must be completed first. guide user to submit-kyc.mjs.

delivery acceptance side effect if user tries to accept delivery, agent must warn: "accepting delivery releases [amount] [currency] to [business] and moves order to rating. this is irreversible except via dispute. proceed?" if user confirms, call accept-delivery.mjs. if user is uncertain, suggest requesting revision first.

deal acceptance side effect if user tries to accept a negotiation proposal with status 'final_deal', agent must warn: "accepting creates a binding contract and order with [business]. all other active negotiations for this intention will be cancelled. [amount] [currency] will be held in escrow until delivery. proceed?" if user confirms, call accept-deal.mjs. if user needs to review other proposals first, stay in negotiation mode.

redescribe intention side effect if user wants to redescribe an intention while negotiations are active, agent must warn: "redescribing will cancel all active negotiations and restart matching. all conversation history with businesses will be lost. continue?" if confirmed, call redescribe-intention.mjs. this is destructive and should only proceed with explicit user intent.

message drafting and confirmation when agent drafts a message for negotiation or order chat, agent must display draft to user and ask for approval before calling send. agent can suggest wording but cannot auto-send without user seeing and confirming it first.

rating generation restriction agent must never generate or guess a user's review rating. user must explicitly provide 1-5 rating. agent can suggest comment text based on delivery quality signals, but rating is user-only.

currency selection on pay-order when calling pay-order.mjs, user must explicitly select USD or USDC. agent cannot assume or default to one. ask "pay with USD or USDC?" before proceeding.

network and rate limit errors if response is 429 (rate limit), agent should wait 10-30 seconds and retry. if response is 500/502/503, inform user that platform is experiencing issues and suggest retry in a few moments. do not cascade errors; stop and ask user if they want to retry.

auth expiry if response is 401 after initial successful call, api key may have expired. ask user to verify key and create a new one if needed.

incomplete business data if get-business.mjs returns sparse or missing fields (e.g., no credit profile yet), inform user that business profile is incomplete and suggest contacting business directly or checking reviews for more info.

output contract

successful operation output formats

• intention created: json object with fields intention_id (uuid), status (string: 'draft'), content (string), locale (string), created_at (iso timestamp)
• intention published: json object with updated status ('published')
• matching triggered: json object with job_id (uuid), status 'pending'
• match status: json object with intention_id, status (string: 'pending'|'running'|'completed'|'failed'), progress_percent (0-100 integer)
• matches retrieved: json array of business objects, each with business_id (uuid), match_score (0.0-1.0 float), business_name (string), reputation (object with avg_rating, review_count, completion_rate), credit_summary (object with score, tier)
• negotiation started: json array of session objects, each with session_id (uuid), business_id (uuid), status ('active'), created_at (iso timestamp), proposal (object or null)
• negotiation message: json object with session_id, sender (string: 'user'|'business'), content (string), timestamp (iso), proposal (object or null if