feat: add Taskmarket skill — on-chain bounty delegation with authorization gates

[YfengJ]

Description

This PR adds a Taskmarket skill — on-chain bounty delegation for Bankr agents.

Taskmarket is a marketplace where agents and humans post digital work escrowed in USDC on Base and workers compete to complete it. With this skill, a Bankr agent gains both sides of the market, driven by the official taskmarket CLI:

Requester flow (mandatory gates built into the skill):

1. Echo the exact task definition — description, reward, deadline, deliverables, network (Base, 8453), and maximum spend (reward + 0.001 USDC relay fee)
2. Require fresh, explicit user authorization in the current conversation before creating or funding anything
3. Create the task (taskmarket task create), escrow on Base, and return the task ID + taskmarket.dev link + escrow tx hash
4. Retrieve live status on demand (task get)
5. Retrieve submissions (task submissions) and present them for human review — never silently accept or reject

Worker flow: browse open bounties, verify escrow from task get before working, and submit with user confirmation.

Safety contract (documented in SKILL.md and references/cli-reference.md, enforced by tests):

• No keys/seeds/tokens ever leave the CLI keystore
• Spend ceiling per task = reward + 0.001; exceeding it is a hard stop
• Never blindly retry a payment with unknown settlement — verify read-only first
• One submission per task; ambiguous submits verified via task my-submissions before retry

Files

File	Purpose
taskmarket/SKILL.md	Skill definition with the full requester/worker workflows and safety gates
taskmarket/catalog.json	Catalog entry following the schema of existing skills (e.g. quicknode)
taskmarket/references/cli-reference.md	Complete CLI command reference
taskmarket/tests/verify_skill.sh	Contract tests: catalog parses, all 6 safety gates present, references resolve
README.md	One row added to the Available Skills table

Tests

$ bash tests/verify_skill.sh
OK: catalog parses, all 6 safety gates present, references resolve

Live demo of the underlying flow (real CLI against production Taskmarket on Base, 2026-08-22):

Setup: npm i -g @lucid-agents/taskmarket@latest && taskmarket init

Prompt: list open Taskmarket tasks with a reward of at least 2 USDC

$ taskmarket task list --status open --reward-min 2 --limit 5
{"ok":true,"data":{"tasks":[
  {"id":"0xace815c5…","reward":"100000000","expiryTime":"2026-08-29T21:40:19Z",...},
  {"id":"0x48e95c92…","reward":"2000000","expiryTime":"2026-08-23T00:16:02Z",...}]}}

Prompt: show me the full details of the second one before any work

$ taskmarket task get 0x48e95c9224caaa8ef99f0931f38143e0a30dfeece12f25a6adaaf3c2c17dacf2
{"ok":true,"data":{"id":"0x48e95c92…","reward":"2000000","status":"open",
 "escrowTxHash":"0x93dd868b…","stakeRequired":false,"submissionWindowOpen":true,…}}

Task creation follows the same gate sequence: echo → explicit approval → create → return ID + link + escrow hash.

Checklist

• Skill folder follows the existing structure (SKILL.md + catalog.json + references/)
• Catalog entry matches schemaVersion 1 used by existing skills
• Listed in the README Available Skills table
• No secrets read, stored, or committed; all paid actions gated on explicit user authorization