Rewrite metamask-agent-wallet skill for weak-LLM execution (v6.0.0)

Full template rewrite of SKILL.md and all references/ and workflows/,
verified against @metamask/agentic-cli 4.0.1 (--help flag schemas +
live read-only runs + compiled-source inspection):

- One canonical syntax per command (flag form wherever both exist);
  removed phantom positionals (predict quote/place/markets get,
  wallet requests watch)
- Required/optional flag tables with value formats; captured real
  output blocks with Capture -> placeholder lines; explicit
  confirm-before-executing checklists; per-file error/recovery tables
- SKILL.md: placeholder legend, global-flags table, preflight
  checklist with real mm doctor output, explicit routing rows
  (added mm tx, price/token subcommands)
- Split predict.md into predict-account/data/trade; market-data.md
  into price/token; renamed chain.md->chains.md,
  polling.md->wallet-requests.md; new concepts.md and aave.md
- New scripts/encode_approve.py for exact-amount ERC-20 approvals
- Fixed: QR login works on prod (COMING_SOON guard is dead code);
  --to-address rejected same-chain (INVALID_SWAP_PARAMS); perps
  --venue optional; --wallet-timeout + walletTimeoutSeconds
  documented; perps --dry-run/--yes coverage; transfer --token
  semantics; Aave ERC-20 vs native value branch

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: yashovardhan

CHANGELOG.md

@@ -10,6 +10,59 @@ catch up if you are on an older skill version — apply the entries above yours
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and the skills follow [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [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.md` → `predict-account.md`, `predict-data.md`, `predict-trade.md`;
+  split `references/market-data.md` → `price.md`, `token.md`; renamed `references/chain.md` →
+  `chains.md` and `references/polling.md` → `wallet-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

README.md

@@ -6,14 +6,14 @@ SKILLs for the MetaMask Agent CLI (`@metamask/agentic-cli` v4.0.1). These skills
 
 | Skill | Description |
 | --- | --- |
-| [`metamask-agent-wallet`](./skills/metamask-agent-wallet/SKILL.md) | Full CLI skill that routes the agent to topic-specific reference docs (`references/`) for all MetaMask Agent CLI commands — auth, wallets, transfers, signing, swaps, bridges, perps, prediction markets, Aave V3, market data, x402 payments, and calldata decoding — plus multistep workflow templates (`workflows/`) for onboarding, swaps, bridges, perps, prediction markets, and Aave. |
+| [`metamask-agent-wallet`](./skills/metamask-agent-wallet/SKILL.md) | Full CLI skill that routes the agent to topic-specific reference docs (`references/`) for all MetaMask Agent CLI commands — auth, wallets, transfers, signing, swaps, bridges, perps, prediction markets, Aave V3, market data, x402 payments, and calldata decoding — plus multistep workflow templates (`workflows/`) for onboarding, swaps, bridges, perps, prediction markets, and Aave. Every doc follows a fixed template (one canonical syntax per command, flag tables verified against `mm --help`, captured outputs, confirmation checklists) so that even small models can execute commands by string substitution. Runtime helper scripts: `amount_to_hex.py`, `encode_approve.py`, `x402_pay.py`. |
 
 ## Installation
 
 Install with [Vercel's Skills CLI](https://skills.sh):
 
 ```bash
-npx skills add metaMask/agent-skills
+npx skills add metamask/agent-skills
 ```
 
 ## Changelog