desk-execution-protocol
Skill418 starsChanged 12 days agoMITView on GitHub
The Execution Trader's procedure for turning an approved ticket into one Hyperliquid action and reconciling it - the pre-send checklist, order construction rules, single-send discipline, unknown-result handling and the execution report. Use before and after every send, cancel, modify, leverage change or close.
2 other projects use a version of this, 1 of them edited.
Adopted by
Most recently taken up by
- swcstudiospace/PumpGroktheir own version, first seen Nov 5, 2026
- Mendurim/strikegrok-trading-deskedited, first seen Sep 10, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 1
- Mendurim/strikegrok-trading-desk★ 018% same
Wrote their own version · 1
- swcstudiospace/PumpGrok★ 42same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository17 other files it gives its agents.
- AGENTS.mdAGENTS.md
- desk-incident-responseSkill
- desk-monitoringSkill
- desk-operating-modelSkill
- desk-post-trade-reviewSkill
- desk-risk-limitsSkill
- desk-strategy-labSkill
- desk-trade-lifecycleSkill
- hypergrok-bootstrapSkill
- hyperliquid-accountSkill
- hyperliquid-advancedSkill
- hyperliquid-api-referenceSkill
- hyperliquid-market-dataSkill
- hyperliquid-ordersSkill
- hyperliquid-positionsSkill
- hyperliquid-setupSkill
- hyperliquid-websocketSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 7 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 23
Run every item and write the result under ## execution before sending.
- Distinctive: line 46
If found: proceed to reconciliation as if the response had arrived, and record that the original response was lost.
- Distinctive: line 47
If not found: a negative check is not proof.
- Very specific: line 61
Write the reconciled facts under ## reconciliation, post the execution report on the floor (format in agents/execution-trader.md), and DM …
- Distinctive: line 85
Never send without a PASS and approval by id for this exact ticket.
- Distinctive: line 86
Never send from a main-wallet key or with a key pasted in chat.
- Distinctive: line 87
Never withdraw, deposit, bridge, transfer, send tokens, approve builder fees or touch vaults and sub-accounts.
- Distinctive: line 88
Never resend on an unknown result while the original could still arrive.
- Distinctive: line 89
Never let a routine, a watch or a schedule send.
- Very specific: line 90
Never treat a clean orderStatus read as proof that an unknown result did not execute.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Execution protocol
This is the only skill on the desk that ends with a request to Hyperliquid's /exchange endpoint, and only the Execution Trader uses it. The API mechanics live in hyperliquid-orders and hyperliquid-positions; this skill is the discipline around them.
Inputs
- The proposal file
/workspace/trading-desk/proposals/<id>.mdwith a Risk Manager PASS and exact ticket fields. - The user's approval line, by id, in chat, after the ticket was shown, inside the ticket's expiry.
- The desk computer configured per
hyperliquid-setup:HYPERLIQUID_NETWORK,HYPERLIQUID_ACCOUNT_ADDRESS, and the API wallet key available to scripts from the environment (never printed).
Pre-send checklist
Run every item and write the result under ## execution before sending. Any failure: do not send, name the item, hand back to the Desk Lead.
- Ticket integrity and gate. Id, PASS and approval refer to the same immutable ticket revision: action, account, network, market, side, size, entry, stop, take-profit, leverage, bounds and expiry. Append revisions; never overwrite the approved text. Any economic change needs a new suffixed ticket id, Risk PASS and user approval; the old approval cannot transfer. Confirm the native out-of-band approval gate covers this send path, including SDKs, scripts and HTTP. If coverage is unverified, do not send, even with chat approval or a written standing scope.
- Network.
HYPERLIQUID_NETWORKequals the ticket's network. Mainnet is never assumed. - Account and wallet.
HYPERLIQUID_ACCOUNT_ADDRESSequals the ticket's account. QueryuserRolefor the public API wallet address on this network: requirerole: agentanddata.userequal to the ticket's account. Check its current approval and expiry withextraAgentsfor that account. An unsigned account read succeeding proves neither signing authority nor wallet scope. If the desk has never sent from this wallet on this network, rehearse through an approved testnet ticket first, not on mainnet. - Price still valid. Fresh mid from
allMidsis within the ticket's slippage tolerance of the ticket price. For a stop or take-profit ticket, the trigger price is on the correct side of the current mark. - Formatting. Asset index from live
meta; price rounded to at most 5 significant figures and at most6 - szDecimalsdecimals for perps (8 - szDecimalsfor spot); size rounded down toszDecimals; notional at least 10 USD; leverage on the account for that market already equals the ticket's leverage (set it first with a separate approved action if not). - cloid, nonce and expiry. Generate a fresh 16-byte client order id (
0x+ 32 hex chars), write it to the proposal file, and put it on the order. One cloid per order, never reused. Record the actual nonce chosen when constructing the signed request, its public API wallet signer and network before dispatch; a nearby send timestamp is not a substitute for the signed nonce. SetexpiresAfteron the send, no later than the ticket's expiry and normally a minute out (hyperliquid-advanced), and write that deadline beside the cloid and nonce. The cloid identifies the order; the deadline closes its future acceptance window. Expiry does not cancel an order accepted earlier, so recovery still requires reconciliation. - One action. Entry plus its stop and take-profit go in one
orderaction withgrouping: normalTpsl(children sized to the entry, placed when it fills). Protection for an existing position is a standalone reduce-only trigger. Anything else in the ticket that is a different action type (leverage change, cancel) is a separate, separately approved step. - Capacity and recovery. Before any action that adds risk or removes protection, no unreconciled send may remain, regardless of its age; complete any restart/feed-gap recovery in
hyperliquid-websocket. Immediately before new risk, have Risk re-read account state, open orders and capacity and recheck the daily stop, limits and all reservations fromdesk-risk-limits. Reserve the approved revision once. Incomplete historical recovery may still permit a reduce-only protective stop or exit when Risk verifies the current position and orders and the ticket can only reduce risk; all other approval, native-gate and single-send checks still apply, and unresolved reservations remain held. A failed applicable gate blocks the send; changed economic fields need a new PASS and approval.
Send
- Send exactly once. Do not wrap the send in a retry loop. Set a finite timeout (10-15 seconds is plenty).
- Capture the raw response and the send timestamp to the millisecond.
- Interpret statuses per
hyperliquid-orders:resting(oid),filled(totalSz, avgPx, oid),waitingForTrigger,waitingForFill, or anerrorstring. A top-levelstatus: erris an action-level rejection: nothing was placed.
Unknown results
A timeout, connection reset, HTTP 5xx, or a client exception after the request left the machine is an unknown result. Treat it as possibly executed:
- Do not resend.
- Query
orderStatusfor the cloid; checkopenOrdersanduserFillsfor it; checkclearinghouseStatefor a position change. - If found: proceed to reconciliation as if the response had arrived, and record that the original response was lost.
- If not found: a negative check is not proof. The original may still be in flight and can land after any number of clean reads. Do not let two quiet checks authorise a replacement.
- Wait for the send's
expiresAfterdeadline to pass, then reconcile once more. The original can no longer be accepted after that deadline, but it may have been accepted beforehand and left orders or fills. Never assume elapsed time without a signed deadline proves the outcome. - If the send carried no expiry, an exact recovery ticket may request a
noopusing the original exact signed nonce, the same API wallet signer and the same network. It requires Risk PASS, user approval by id and native-gate coverage; the Execution Trader sends it once. A fresh higher nonce does not invalidate the original. Confirm that the noop was accepted, then reconcile the original and account; the original could have won the race. A nonce-reuse rejection does not confirm noop acceptance or identify what consumed the nonce. If the original nonce is missing or the noop result is unconfirmed, keep the unknown send reserved, pause new risk and escalate to the Desk Lead and user. Do not improvise another signed action. - Report the acceptance deadline or confirmed nonce invalidation and the reconciled outcome separately. Only after the original cannot be accepted later and its existing exposure is reconciled can a replacement proceed on a fresh ticket, Risk PASS and approval by id.
Reconciliation
Immediately after the response, and again when fills arrive:
orderStatusby cloid or oid: the exchange's view of the order.openOrders/frontendOpenOrders: resting orders including trigger children.userFills(oruserFillsByTimefor the window): fill price, size, fee,crossed(taker or maker), timestamp.clearinghouseState: position size, entry, leverage, liquidation price, margin used.
Write the reconciled facts under ## reconciliation, post the execution report on the floor (format in agents/execution-trader.md), and DM the Trade Reviewer with the id and the report.
Update the risk reservation with those same facts. A resting remainder or unknown send retains its allocation; a cancellation response alone does not free it. Filled exposure moves into the position calculation, and only reconciled unused capacity is released. Record the change so a restarted desk can reconstruct it from proposals and the exchange.
Cancels, modifies, leverage, closes
Each is its own ticket (suffix -B, -C...) with its own PASS and approval by id, unless the user has written a standing approval into desk.md for that exact class of action on that network.
- Cancel: by cloid where you have it, else by oid from
openOrders. Confirm removal fromopenOrders. - Modify: Hyperliquid's
modify/batchModifycancels a resting limit order and places the replacement in one action (new oid, fresh cloid); the replacement must itself rest (Alo, or a non-executableGtc). Stops and take-profits cannot be modified: place the new trigger order, confirm it is resting, then cancel the old one, so the position is never unprotected. - Leverage or margin mode:
updateLeveragebefore the entry, on a flat market or with the user aware of the effect on an existing position. Confirm fromclearinghouseState. - Close: reduce-only IOC at a slippage-bounded price for the position size read live seconds before the send. Then read
clearinghouseStateagain: if the position is gone, cancel the orphaned protective orders; if it only shrank (a partial fill on the close), the remainder is still open and still needs its stop, so resize protection before cancelling anything. Report which case it was. - Dead-man's switch:
scheduleCancelonly when the user asks for it, with the time written in the report; remember it cancels all of the account's open orders when it fires.
Rehearsal rule
Any action type the desk has not performed before (first TP/SL bracket, first modify, first close, first isolated-margin trade) is rehearsed on testnet with the same ticket format before it is done on mainnet. Record the rehearsal in the journal.
Report
Post the block from agents/execution-trader.md on the floor: sent, response, reconciled, fees/funding, next. Keep the raw response in the proposal file, not in chat.
Never
- Never send without a PASS and approval by id for this exact ticket.
- Never send from a main-wallet key or with a key pasted in chat.
- Never withdraw, deposit, bridge, transfer, send tokens, approve builder fees or touch vaults and sub-accounts.
- Never resend on an unknown result while the original could still arrive; only an expired original and a fresh approval by id permit a replacement. Never cancel-all as a reflex.
- Never let a routine, a watch or a schedule send. They alert and draft; only the Execution Trader sends, and only on an approved ticket.
- Never treat a clean
orderStatusread as proof that an unknown result did not execute; only an expired original proves that.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

