Skip to content

Latest commit

 

History

History
118 lines (91 loc) · 6.6 KB

File metadata and controls

118 lines (91 loc) · 6.6 KB

Changelog

All notable changes to the MetaMask Agent CLI Skills are documented here.

These skills track the @metamask/agentic-cli releases. Each entry lists the skill version (frontmatter) and the CLI cliVersion it targets, along with the user-facing CLI changes that motivated the skill update. Use this log to catch up if you are on an older skill version — apply the entries above yours in order.

The format is based on Keep a Changelog, and the skills follow Semantic Versioning.

[6.0.0] — targets CLI v4.0.1

Changed (breaking)

  • Full rewrite of SKILL.md and every file in references/ and workflows/ to fixed, weak-LLM-friendly templates: one canonical syntax per command (flag form wherever both positional and flag forms exist), required/optional flag tables with value-format columns, captured real output blocks with Capture: field → <placeholder> lines, explicit "Confirm before executing" field checklists, per-file error/recovery tables, and a placeholder legend + global-flags table centralized in SKILL.md. Every documented flag was verified against mm <command> --help JSON on CLI v4.0.1. Incremental catch-up from ≤5.x does not apply to this entry — re-read the whole skill.
  • Split references/predict.mdpredict-account.md, predict-data.md, predict-trade.md; split references/market-data.mdprice.md, token.md; renamed references/chain.mdchains.md and references/polling.mdwallet-requests.md.

Added

  • references/concepts.md — shared concepts: wallet/trading modes, async job model (pollingId, --wait, intent), secrets via MM_PASSWORD/MM_MNEMONIC, $SKILL_DIR, CAIP-2/CAIP-19 construction, amount/decimals conversion, suspicious-payload checklist.
  • references/aave.md — canonical home for the Aave V3 GraphQL machinery (endpoint, market discovery, amount formats per operation, the three response types, approval security rule).
  • scripts/encode_approve.py — deterministic ERC-20 approve(address,uint256) calldata for exact-amount allowances (replaces the old hand-encoding instruction in the supply workflow).

Fixed

  • Removed phantom positional syntax: predict quote/place/markets get and wallet requests watch are flag-only (--token-id, --market, --polling-id).
  • predict redeem: exactly one of <condition-id> | --all (previously shown as both optional).
  • predict markets search takes a positional query (mm predict markets search <query>).
  • Documented --wallet-timeout on all wallet-job commands and the walletTimeoutSeconds config key; MM_PASSWORD env applies to every command that accepts --password.
  • perps: --venue is optional (defaults to hyperliquid); --dry-run/--yes documented on cancel/transfer/deposit/withdraw (the old "--yes has no effect" claim was false).
  • transfer --token: ERC-20 transfers use the contract address; symbols only for native.
  • QR login (mm login qr) works on all environments including production — the COMING_SOON-on-prod claim in the onboarding workflow was stale (verified against CLI source).
  • --to-address is rejected for same-chain swaps (INVALID_SWAP_PARAMS) — documented with the verbatim error; --to-address/--refuel are cross-chain only.
  • Aave supply: explicit ERC-20 ("value":"0x0") vs native (hex from amount_to_hex.py) branch; per-operation amount formats consolidated in references/aave.md.
  • SKILL.md routing: added mm tx (lookup by hash); replaced the vague mm price ... / mm token ... rows with one row per subcommand.

Notes

  • The CLI's own --help usage strings are unreliable in places (e.g. mm wallet trading-mode get --help prints mm mode get, which does not exist, and advertises --chain-namespace/--address flags that are rejected). The docs trust the flags arrays and empirical runs instead; upstream bug to file against MetaMask/agentic.

[5.0.1] — targets CLI v4.0.1

Removed

  • metamask-agent-workflows skill. Its workflow templates were a subset of the metamask-agent-wallet skill's workflows/ directory (the wallet copies were also more up-to-date). The repo now ships a single metamask-agent-wallet skill that bundles both the reference docs and the workflow templates. Install it with npx skills add metaMask/agent-skills.

Changed

  • references/doctor.md: document that mm doctor now detects project-local MetaMask AI skills when the global skill lock file (~/.agents/.skill-lock.json) exists but lacks MetaMask entries. (agentic#263)
  • Bumped cliVersion to 4.0.1 across both skills and updated README.

Notes

  • CLI v4.0.1 also improved Sentry error reporting and surfaced reportable failures in Segment analytics (agentic#264); these are internal telemetry changes with no impact on the command surface, so no skill docs changed.

[5.0.0] — targets CLI v4.0.0

Added

  • BYOK (bring-your-own-key) parity for mm init and mm wallet create: persisted BYOK wallets are registered server-side, prompt for a trading mode, and apply the server trading policy — matching the server-wallet flow. (agentic#246)

Fixed

  • mm predict deposit-wallet setup no longer stalls with contradictory relayer errors when a cached address was derived as legacy UUPS instead of BeaconProxy. (agentic#259)

Earlier CLI history

For CLI releases prior to v4.0.0, see the upstream packages/agentic-cli/CHANGELOG.md. Highlights that shaped the current skill surface:

  • v3.2.0 — confirmation prompts (with --yes) before mm logout / mm reset; npm update notifier; mm wallet list refreshes the remote roster before listing.
  • v3.0.0breaking: mm login google / mm login email removed; use mm login browser (Google or email via the MetaMask dashboard). Bridge --refuel added; deposit preflight checks; auto-rehydrate on login.
  • v2.0.0breaking: wallet policy moved to mm wallet policy get|set|template (mm wallet show / create return policyYaml instead of policies). Added mm doctor, mm transaction history, and major CLI startup performance work.
  • v1.0.0mm wallet add-fund (QR funding), trading-mode get/set commands, and mm swap/mm predict recipient support.