Candle ships 23 skills: plain-text instructions an agent reads before it acts. They teach an agent what the API surface will not: when to trade, how to size, what to check before a buy, and how to keep an unattended bot safe. Install them with the Candle plugin or CLI; see Coding agents.
A skill never holds a key and never moves funds by itself. The agent still works through a scoped key; see Agents and keys.
The 23 skills by name: candle-agent-operations, candle-book-limits, candle-drained-pool-writeoff, candle-first-live-trade, candle-kill-switches, candle-launch, candle-llm-trade-judge, candle-market, candle-market-data-ingestion, candle-risk-gate, candle-setup, candle-strategy-design, candle-strategy-hunt, candle-strategy-lab, candle-strategy-to-bot, candle-trade, candle-trade-execution, candle-trading-discipline, candle-transfer-fee-tokens, candle-wallet-entry-detection, candle-wallet-layout, candle-webhooks, onchain-strategy-research.
| Skill | What it teaches |
|---|
candle-market-data-ingestion | Consume hacc event feeds and Solana wallet or pool activity without silently losing events: durable cursors, receipt times, websocket plus reconciliation, transaction versions, gaps and pruning. Use when building or debugging a collector, a wallet watcher or a signal feed, when a feed seems to have gone quiet, and when signals appear late or not at all. |
candle-wallet-entry-detection | Decide whether a watched wallet actually bought a token, whether it still holds it, and which of its moves should count toward a signal or an exit. Use when building a wallet-following or wallet-convergence signal, when counting how many tracked wallets entered a token, and when a follow signal fires on wallets that did not really buy or no longer hold. |
| Skill | What it teaches |
|---|
candle-webhooks | Register a webhook endpoint and verify signed Candle event deliveries for launches, graduations, migrations, trades, and limit-order triggers. Use when the user asks about webhooks, notifications, or reacting to fills, graduations, or triggered orders instead of polling. |
| Skill | What it teaches |
|---|
candle-trade | Buy or sell a Candle-launched token, or place a Max-tier limit order that triggers server-side and fills once your agent comes online to complete it. Use when the user asks to buy, sell, swap, trade, or set a limit order. |
| Skill | What it teaches |
|---|
candle-trade-execution | Use when building or debugging automated trading on the Candle agent SDK (@candledottv/agent-sdk), especially position accounting, sells that fail, retry loops, and error handling. Encodes failures that cost real money in production. |
onchain-strategy-research | Research onchain trading mechanics, audit wallet and token evidence, and turn observations into reproducible strategy experiments with executable fills, temporal validation, and capital controls. Use for onchain strategy research and trading-pipeline reviews. |
| Skill | What it teaches |
|---|
candle-market | Read a Candle token’s market state and lifecycle, curated feeds carrying live price and market cap (new, graduated, onfire, and finalstretch where it is switched on), and public agent profiles, with no API key and no signup. Use when the user asks to check a token’s market state, browse trending or graduated tokens by price or market cap, or look up an agent’s public profile. |
| Skill | What it teaches |
|---|
candle-agent-operations | Run long-lived trading workers and data collectors so they recover from crashes without losing state, never extend their own run, report health truthfully, and hand over cleanly to the next agent session. Use when deploying a trading worker or collector that runs for hours or days, when a worker has crashed, hung or restarted, and when ending a session that leaves workers running. |
candle-first-live-trade | The plumbing test a trading bot needs before and during its first live trades: a supervised minimum-size round trip through the real code, a sell proven before buys scale, a control exercised on a live position, and the ledger reconciled against the wallet. Use when a bot is about to trade real money for the first time, after any change of key, signer, wallet, venue or quote asset, when an operator says to kill a newly live bot, when the first live fills look different from paper, and when a newly live book goes quiet. |
candle-strategy-to-bot | Build a trading strategy spec into a deterministic bot: policy as data with approval bound to its exact content, variants as settings with parity tests, one code path for replay, paper and live, an explicit position state machine, a ledger of every decision including refusals, fail-closed entries that never block exits, and the test order before money. Use when a strategy spec must become a bot, or when a running bot’s rules must change. |
| Skill | What it teaches |
|---|
candle-llm-trade-judge | Test whether a language model adds anything to trading decisions before trusting it, and give it the role where that can be measured: a judge or veto over a deterministic entry rule. Use when putting a language model in a trading loop, when evaluating an agent that decides entries or exits, and when a model-driven strategy keeps converging on the same trades whatever the prompt says. |
candle-strategy-hunt | Find trading leads worth testing and triage them cheaply before full validation: start from what the market just did, define a census universe with a random control, name who pays, know the regime, and run the cheapest decisive test first under your own costs and latency. Use when asked to find something to trade, when there is no lead yet, or when a raw idea needs a decision on whether it deserves a test. |
candle-strategy-lab | Replay a strategy definition against recorded data in the hacc strategy lab and read the lead card honestly: beat the random controls first, then break-even above your real cost, then the regime. Use when a strategy idea or definition needs a measured backtest, or to read the lab’s runners and edges tables. The lab never trades. |
| Skill | What it teaches |
|---|
candle-book-limits | Size and audit the lifetime caps, slot counts and wallet reserves of an autonomous trading book so they bound risk without silently stopping the book or blocking its exits. Use when designing a book’s limits, when a running book has gone quiet, when an order is stuck unresolved, and when two books share one wallet. |
candle-drained-pool-writeoff | Recognize a held position whose pool was emptied after entry, and account for it as a loss instead of marking it at a stale price or retrying sells forever. Use when a held token’s price has stopped updating, when sell quotes keep failing, when a mark looks healthy but no trades arrive, and when deciding whether an unsellable position can be closed out. |
candle-kill-switches | The operator controls an autonomous trading book needs so a human can stop new risk, cut one position or cut everything without editing code or killing a process mid-order, and how those controls differ from the rail’s own profile stops. Use when building a trading worker, before running one with real money, and when an operator asks to pause, cut, stop or restart a running book. |
candle-risk-gate | Decide whether a token is safe enough to buy, make sure the position carries its own exit before you own it, and judge whether a wallet is worth following. Use before any buy, before copying any trader, and whenever a check comes back unavailable rather than clean. |
candle-trading-discipline | How much to bet, when to stop, and the arithmetic that decides whether a strategy can be profitable at all. Use before running any strategy with real money, when sizing a position, when deciding whether to keep going after losses, and whenever somebody asks an agent to trade until it profits. |
candle-transfer-fee-tokens | Judge the real round-trip cost of a Token-2022 mint with a transfer fee without double-counting or rejecting it by mistake, size and sell from what was actually received, and reconcile paper against live correctly for taxed tokens. Use before buying a token whose mint carries a transfer-fee extension, when every taxed token fails an impact check, when a new position marks well below entry immediately, and when paper and live results disagree on taxed tokens. |
| Skill | What it teaches |
|---|
candle-setup | Install the Candle CLI, authorize a device from the browser, set up the self-custody vault and a TEE wallet for an agent, and provision an agent API key at the right access level. Use when the user asks to install, log in, authenticate, configure credentials, set up a wallet for an agent, or connect their Candle account. |
candle-wallet-layout | How to lay out agent keys, TEE wallets and key signers when running more than one strategy, what is shared between wallets on one key, and how to move funds without landing on a look-alike address. Use when giving a second strategy its own wallet, choosing between one key and a key per bot, setting up a key signer on a trading machine, rebinding a wallet to another key, cleaning up old keys, sending funds between your own wallets, and whenever a destination address would come from transaction history. |
| Skill | What it teaches |
|---|
candle-strategy-design | Turn a trading lead into a strategy that can be tested, run and judged: a one-page spec, judging and sizing matched to the payoff shape, exits designed per profit source, costs and break-even, a bar written before live, layers judged against the layer below, paired variants, running without interference, and a close-out verdict. Use when a lead must become a strategy, or when a running strategy needs amending or judging. |
| Skill | What it teaches |
|---|
candle-launch | Launch a new token on Candle, on Solana or Hood, optionally seeded with a dev buy (same transaction on Solana; best-effort follow-up on Hood). Use when the user asks to launch, create, deploy, or seed a token. |
