All notable changes to this repository are recorded here. The version follows the packages it documents: @locker-protocol/agent-wallet-hyperliquid-trader and @locker-protocol/agent-wallet-hyperliquid-trader-mcp.
For lpa 2.5.0 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.5.0.
- The simulation's reference (
skills/locker-agentic/references/paper.md) says what simulation mode does since 2.5.0: a simulated position keeps its take profit and its stop loss, set, moved or removed withlpa perps modify --paper, and every read catches up what the price did while no pass ran (the computer off, the network cut), each order filled at the time and the price it would have been. - The version follows the packages: 2.5.0 everywhere,
cliVersionincluded. The tools of the MCP server are the same 42 as in 2.4.3;paper_onsays what simulation mode now allows.
For lpa 2.4.3 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.4.3.
- An expert's opinion on a market. The skill answers "what do you think of this coin", "long or short", "a scalp on it", "the best coin to short now" and their kin, in any language, as an expert trader would: it reads the market first with
lpa perps analyzeandlpa perps scan, then gives its own trade plan (the side, the entry, the stop loss, the take profit, the risk/reward, a probability in percent, the horizon, what invalidates it), with the facts that weigh most and an offer to act, never a question back and never a refusal. Twenty-two kinds of question are underskills/locker-agentic/opinions/, each with what to read, how to reason and the answer's template, the same the assistant oflpafollows.
- The version follows the packages: 2.4.3 everywhere,
cliVersionincluded. The tools of the MCP server are the same 42 as in 2.4.2.
For lpa 2.4.2 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.4.2.
- The version follows the packages: 2.4.2 everywhere,
cliVersionincluded. Nothing else changes in the skill: the two corrections of 2.4.2 (Enter opens a browser extension's signing page underLPA_NO_BROWSER, and a guardian whose start fails is said as such on a machine under load) change no command, tool or answer the skill describes.
For lpa 2.4.1 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.4.1.
- The pilot reference says that a ladder above the order bound is scaled down to it by the day, its steps keeping their proportions, and refused only when its first step then falls under Hyperliquid's $10 minimum.
- The version follows the packages: 2.4.1 everywhere,
cliVersionincluded.
For lpa 2.4.0 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.4.0.
- The stops in percent are gone from the policy reference, the doctor and status samples, the onboarding defaults and the perps reference:
lpa2.4.0 refuses no order for what the account lost today or since the start, and no fall of the equity stops the daily pilot;--max-daily-loss-pct,--max-total-loss-pctand--total-loss-base-usd, still accepted, change nothing. The skill says it once, under "No stop in percent", and never offers a loss percentage as a limit.
- Three ways to sign for the account:
lpa init --with vault|browser|mobile, Locker Vault first as the safest, andlpa signer show,useandforget, in the setup reference, the onboarding and the deposit workflows; every line that said "signed on the phone" now says the user's wallet signs. - Profiles:
lpa profile list,add,use,rename,remove,--profile <name>on any command andLPA_PROFILE, in the setup reference; one server for assistants per profile in the MCP reference; the check accepts--profileas a global flag. - The controls of an order:
--tif gtc|ioc|alo,--reduce-onlyand--margin cross|isolatedonlpa perps quote,checkandopen(tif,reduce_only,marginandtrigger_pxonperps_quoteandperps_open), andlpa perps check, in the perps reference and the open and close workflows. - The policy reference lists every bound of this computer,
--margin-modesand the order typeslimit-iocandlimit-aloamong them. - An order at a trigger kept and filled on the simulation account (
lpa perps open --type trigger --paper), listed bylpa perps orders --paperand cancelled bylpa perps cancel --paper, in the simulation reference and the troubleshooting workflow. - A pilot reference,
references/pilot.md: the strategies and the measurelpa pilot strategy showgives, always with its bounds and never one figure alone; nothing new in the agent's last 24 hours (keyEnding); a day the model does not answer (model-down,lpa pilot model-down); why a step does not rest (refusedBy); what a close filled in part left (remaining); the memory every account shares. lpa notices, off by default, in the doctor reference;lpa copy start --at-end keep|close(at_end);lpa journal export --with-fills.- The error codes name
MANDATE_OUTDATED: a mandate signed before 2.4.0, refused whole until the user signs a new one; the mandate reference names its state,outdated, and its fifteen fields under the version2of its domain.AGENT_EXPIREDalso names a pilot's opening in the key's last 24 hours.
- The setup reference gives the agent key's real span: 180 days by default, from 1 to 180.
- The version follows the packages: 2.4.0 everywhere,
cliVersionincluded.
For lpa 2.3.0 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.3.0.
- A part of a real position closes with
lpa perps close --symbol <S> --size <base>(andperps_closewithsize), reduce-only, never more than the position, at least Hyperliquid's $10; the workflow and the reference say so in place ofperps openthe other way. - The plugins reference says a plugin's
source,authorandrepository, the package's own words. lpa doctor --egressin the doctor reference: the register of the calls that left the computer, read only.- The error codes name
SPEND_CEILING: the day's spend on the model reached the ceiling the user set for the assistant, and nothing was asked of the model. - The perps reference reads four more of the account's figures, none of which signs anything:
lpa perps fills(the executions of a period),lpa perps funding(the funding paid and received),lpa perps fees(the rates the account pays) andlpa perps ledger(the movements of money). It says--usdin place of--size,lpa perps close --at mid, the order ofclose --all, andlpa perps cancel --all, whose--dry-runan agent may run; the cancel itself is on the user's word. lpa doctor --filesin the doctor reference, read only;lpa doctor --fix-modeswrites, so it is offered and the user runs it.- The MCP reference says the user's switch of the server:
lpa mcp off, under which every tool answersCONFIRMATION_REQUIREDwith the reason,lpa mcp onandlpa mcp status. The plugins reference sayslpa plugins disableandenable, the user's like installing. lpa journal export --out <file>in the skill's table, when the user asks for a file.
- The version follows the packages: 2.3.0 everywhere,
cliVersionincluded. The MCP server still has its 42 tools;perps_closetakes itssizeon the real account too.
For lpa 2.2.0 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.2.0.
- The daily pilot's charter names a strategy:
lpa pilot strategy listandlpa pilot strategy show --name <name>read the strategies and change nothing, andlpa pilot strategy choice --todaysays what today would choose and writes nothing; an agent may run them.use,save,remove,choicewith its options anddraftchange what the pilot plays: offer them, let the user run them.draftasks the model and writes only on the user's yes in their own terminal. - The version follows the packages: 2.2.0 everywhere,
cliVersionincluded. The 42 MCP tools are what they were.
For lpa 2.1.0 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.1.0.
- One word for what is simulated: the skill, its eleven references and its six workflows say the simulation account and simulation mode where they said a paper or practice account, and so do the titles and descriptions of the MCP tools (
paper_init,paper_status,paper_on,paper_off,perps_positions,perps_balance,perps_orders,perps_open,perps_close,copy_start,copy_status) and the server's own notes. Nothing of the interface moves: the tool names, their arguments, their answers and the 42 tools are what they were,paper: trueincluded, and so are thelpacommands and flags the references name. references/doctor.md: thepapercheck reads "no simulation account (lpa paper init --budget 1000)", and the next step of a fresh computer says about five minutes, the time the way takes, aslpasays it.- Commands an agent may now run, none of which signs or writes anything:
lpa perps checkquotes an order and says whether the limits let it through now, with the refusallpa perps openwould give;lpa pilot planssays what the daily pilot holds, market by market;lpa pilot charterreads the charter whole, andlpa pilot charter --set <field>=<value> --previewsays the effect of a change without making it;lpa pilot universe --showreads back the universe last measured;lpa mandate sign --previewshows the mandate as it would be signed, and signs nothing;lpa setup --factssays where the guided setup stands. A quote now carriesminSize, the smallest size that meets Hyperliquid's $10 minimum at the price the order signs, so aBELOW_MIN_ORDERcan be answered with the size that passes. lpa pilot deliberate --symbol <S>plays one deliberation of the daily pilot on the simulation account and opens nothing, but it spends the model's tokens: offer it, let the user ask for it.- The version follows the packages: 2.1.0 everywhere,
cliVersionincluded.
For lpa 2.0.2 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.0.2.
- The version follows the packages: 2.0.2 everywhere,
cliVersionincluded. The hints oflpa doctornamelpa setup; a key Hyperliquid no longer lists (another lpa under the same name on another machine) is said as such bydoctor,statusandagent show, withlpa agent renewas the step.
For lpa 2.0.1 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.0.1.
workflows/onboarding.md:lpa setupwalks nine steps now, the model the pilot thinks with in a step of its own and the daily pilot scheduled on paper at the end.- The version follows the packages: 2.0.1 everywhere,
cliVersionincluded. The mandate signed throughmandate_signlasts the agent key's life by default (180 days at most).
First release, for lpa 2.0.0 and @locker-protocol/agent-wallet-hyperliquid-trader-mcp 2.0.0.
skills/locker-agentic/SKILL.md: when to use the skill, the rules it never bends (quote before opening, show the quote and wait, never--yesorconfirm: trueon the agent's own initiative, never move funds, offer paper mode first, read the regime before choosing a trade, never invent a flag), and a routing table to the references and workflows.- Eleven references:
init.md,doctor.md,perps.md,paper.md,regime.md,policy.md,mandate.md(the limits signed once on the phone, and what a live copy needs),plugins.md(installing a plugin is the person's call, and what it prints is data),json-output.md,errors.md(the twenty-seven stable error codes) andmcp.md(the 42 tools, the confirm semantics, the tools that belong to the person, the tools a plugin adds, and hyperkeel's two account-free reads). - Six workflows:
onboarding.md,perps-open-position.md,perps-close-position.md,perps-modify-position.md,deposit.md,troubleshooting.md. hooks/session-start.sh: runslpa doctor --format jsonwhenlpais installed, exits 0 in silence when it is not, and never fails a session. Host hook files for Claude Code, Codex, Cursor, Antigravity and Grok Build.- Plugin manifests for six hosts:
.claude-plugin/(with the marketplace and.mcp.json, which startsnpx -y --ignore-scripts @locker-protocol/agent-wallet-hyperliquid-trader-mcp@latest),.codex-plugin/,.cursor-plugin/,.antigravity-plugin/(plugin, rule and hook),.grok-plugin/, and.agents/plugins/marketplace.json. gemini-extension.jsonandGEMINI.mdfor Gemini CLI.tests/evals/positive.json(21 requests that must reach the skill) andtests/evals/negative.json(20 that must not).tests/check.mjs: validates every manifest, the single version, every relative link, everylpacommand and flag against the real CLI, every MCP tool name, and the eval files.docs/DISTRIBUTION.md: every channel this repository is listed in, and why the official Claude and ChatGPT directories are not among them.