Compound III borrowing
Compound III borrowing workflow and evidence
Galleon Labs
@galleonlabs
Install
$ openclaw skills install @galleonlabs/galleon-compound-borrowCompound III borrowing
Work on one Comet deployment, not Compound V2 cTokens. Resolve chain, Comet proxy, base token and account. Compound III borrows one base asset per market; a supplied collateral asset does not earn the base supply interest rate.
Read the position
Read baseToken(), base decimals, baseBorrowMin(), balanceOf(account), borrowBalanceOf(account), numAssets(), and each getAssetInfo(index). For each collateral, read collateralBalanceOf(account, asset), its scale, priceFeed, borrowCollateralFactor, liquidateCollateralFactor and supplyCap. Read prices through Comet's getPrice(feed) and retain their scales. Do not use an unrelated market's USDC configuration for a WETH base market.
Read isBorrowCollateralized(account) and isLiquidatable(account), plus supply/withdraw pause flags. Inspect utilization and rates through the deployment's documented rate functions; APR estimates must disclose accrual interval and exclude rewards unless valued separately. Check available base liquidity independently from collateral value.
Construct the action
Compound III uses withdraw(baseToken, amount) to remove base supply and then create debt if the withdrawal exceeds the positive base balance. If supply is 400 and withdrawal is 1,000, the new debt is about 600 before accrual, not 1,000. A new borrow must satisfy the configured minimum position size. Do not increase an amount merely to meet baseBorrowMin without authorization for the larger debt.
Calculate two limits separately: borrow capacity from borrowCollateralFactor and liquidation capacity from liquidateCollateralFactor. Passing liquidation capacity is insufficient for a new borrow. Simulate the actual withdrawal to handle accrued indices, rounding and configuration exactly; an offchain budget alone is not acceptance.
Repaying uses supply(baseToken, amount). It reduces debt first; excess becomes base supply. Collateral supply is different and does not repay debt. An ERC20 approval permits supply, while Comet manager allow grants account-management authority. A direct self-action should not silently introduce a manager or Bulker approval.
For withdrawal of collateral, model remaining collateral and debt at current and stressed oracle prices. An LTV-style estimate cannot substitute for both protocol boolean checks. After execution, reread base supply, debt and collateral individually; a successful approval or supply event is insufficient to prove a requested borrow or full debt repayment.
Evidence and completion
Return the requested decision, exact deployment and chain, block/time, raw-unit observations, calculation assumptions, and the next missing read. Distinguish an indexed estimate, an onchain read, a simulated call and a confirmed transaction. Research ends with an actionable decision record; it does not require connecting a wallet.
For a requested write, prepare the complete unsigned sequence and simulate with the actual sender, amount, recipient and allowance state. Preserve authorization already supplied; research does not grant debt, spending or signing authority. Never request keys. After an authorized transaction, verify its receipt and the relevant balance or position change. On an ambiguous timeout, reconcile its hash and nonce before retrying.
Read protocol references and worked record only for the selected operation. Documentation checked 2026-09-27; refresh deployments and current parameters at use time. This directory is independently installable.
Top skills in this category
Skill Vetter
@spclaudehomeSecurity-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
Github
@steipeteInteract with GitHub using the `gh` CLI. Use `gh issue`, `gh pr`, `gh run`, and `gh api` for issues, PRs, CI runs, and advanced queries.
Humanizer
@biostartechnologyRemove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases.
Free Ride - Unlimited free AI
@shaivpidadiManages free AI models from OpenRouter for OpenClaw. Automatically ranks models by quality, configures fallbacks for rate-limit handling, and updates opencla...
Elite Longterm Memory
@nextfrontierbuildsUltimate AI agent memory system for Cursor, Claude, ChatGPT & Copilot. WAL protocol + vector search + git-notes + cloud backup. Never lose context again. Vibe-coding ready.