Daily Pulse Trading Workflow
One command to run the daily portfolio review / monitor / rebalance loop for the predict-raven Polymarket trading agent — real-money by default, or recommendation-only on request.
The skill defaults to real-money order execution (AUTOPOLY_EXECUTION_MODE=live) without an explicit confirmation step, a dangerous default with high external effects. Although execution semantics and artifact paths are disclosed, least-privilege scoping, capital limits, rollback/abort mechanisms and failure recovery are absent, so heavy deductions apply.
The command chain maps to package. (daily:forecast → scripts/daily-pulse.ts); instructions are self-consistent and variants are clear — a plausible happy path. But undeclared dependencies (.env.pizza, Redis/BullMQ, keys), missing error-handling documentation, and CI-baselined type errors in the live entry scripts (non-blocking typecheck) keep this at/below the static cap of 10.
Trigger phrases and outputs are clearly described with bilingual docs, but capability/non-fit boundaries are barely declared, and core function depends entirely on overseas services (Polymarket, Exa/Tavily), posing real reachability issues from mainland China — disclosed in cautions.
Reasonable layering (SKILL.md / EN / agents yaml), consistent name-vs-capability, MIT license and CI present at repo level; but the skill itself lacks versioning/changelog, maintenance ownership and update path, and the publisher is unverified.
Claims review/monitor/rebalance reports and live execution, but the README states the live loop is paused (last run 2026-06-10) and discloses a search-evidence contamination incident; output quality and direct usability are unverifiable in static review and comparative-benefit evidence is thin, so well below the static cap of 7.
Evidence is author claims and code organization only; despite repo-level CI and test-suite descriptions, no committed tests covering this skill's key path (live daily pulse) or independent execution evidence exist, and key claims (trades, guardrail blocks) rest on external web pages — low score under the static cap of 5.
- Dangerous default: real-money order execution is the default route; --recommend-only is used only when explicitly requested. Always confirm mode and wallet environment before running.
- Requires private-key/.env credentials and connects to Polymarket mainnet — real capital-loss risk; the live loop has been paused since 2026-06-10, so actual availability is questionable.
- Core function depends entirely on overseas services (Polymarket, Exa/Tavily, forecasting-agent.com), with limited reachability from mainland-China networks.
- The repo discloses a search-evidence contamination incident (DDG CAPTCHA pages parsed as empty results); historical reports are affected and the post-fix audit is incomplete.
- Static review only, nothing executed; no committed test evidence covers the skill's key path, so output correctness cannot be verified.
What does this skill do, and when should you use it?
daily-pulse is a skill inside the predict-raven repository — the framework behind the first autonomous, continuously-running trading agent on Polymarket. Its single entry point, `pnpm daily:forecast`, runs a full cycle of Market Pulse research, portfolio review, monitoring, and rebalancing, archiving every recommendation or executed order. By default it runs the live real-money path against the `.env.pizza` test-wallet environment; it only falls back to `--recommend-only` when you explicitly ask for a preview. It is aimed at traders who have already installed the repo and configured a wallet.
Runs pnpm daily:forecast, which internally invokes forecast:live with ENV_FILE=.env.pizza, AGENT_DECISION_STRATEGY=pulse-direct, and AUTOPOLY_EXECUTION_MODE=live. It fetches the Polymarket Pulse candidate pool, generates a TradeDecisionSet via the pulse-direct engine (ranks by monthlyReturn, takes top 4, caps a round at 20% of bankroll), and pushes it through service-layer hard risk controls (FOK market orders, ≤15% per trade, ≤80% total exposure, ≤30% per event, NAV-drawdown ≥30% global halt) before executing or just producing recommendations. Artifacts land in runtime-artifacts/pulse-live/<timestamp>-<runId>/ and runtime-artifacts/reports/{pulse,review,monitor,rebalance}/, and the run should report the env used, execute vs recommend-only, runId, archive paths, plus new opens, suggested reductions, and guardrail-blocked items.
- A trader with a configured Polymarket test wallet in predict-raven who wants a one-command daily live rebalance
- An unfunded newcomer running `--recommend-only --` to validate the full pipeline without placing orders
- A user with an existing pulse artifact re-running quickly via `--pulse-/--pulse-markdown` instead of re-researching
- An operator or auditor checking daily runIds, archive directories, and guardrail-blocked orders
- A trader re-evaluating the same Pulse under the legacy provider-runtime decision strategy
What are this skill's strengths and limitations?
- One command closes the full daily loop: research, decision, execution, archival
- Risk controls are service-layer hard rules (FOK, position/exposure caps, drawdown halt) that agent errors cannot bypass
- Every decision and artifact is archived with a runId, making runs auditable
- Recommendation-only mode lets you run end-to-end with zero funds at risk
- Defaults to real-money execution — a mis-triggered run spends actual funds, risky for beginners
- Deeply coupled to the predict-raven repo structure (.env.pizza, vendor sync, pnpm); not usable standalone
- No verifiable performance/ROI data in the source; only links to a decision-log website
- README is dated 2026-06-12 and evolves quickly; verify current repo state before relying on it
How do you install this skill?
This skill lives at skills/daily-pulse/ inside the predict-raven monorepo and cannot be installed standalone. Prerequisites: install Claude Code or Codex CLI, git clone https://github.com/Alchemist-X/predict-raven, then tell the Agent inside the repo to 'install the dependencies for predict-raven' (runs pnpm install + pnpm build). For the live path, configure a wallet env file per the README (e.g. .env.pizza or .env.live-test) and run pnpm vendor:sync for the vendor repos the pulse/live paths require. Finer skill-specific installation steps are not documented in the source.
How do you use this skill?
In a configured repo, tell the Agent 'run the daily pulse' or ask for a portfolio review / monitor / rebalance round; it executes pnpm daily:forecast. Common variants: pnpm daily:forecast -- -- for JSON output; pnpm daily:forecast -- --recommend-only -- for suggestions without orders; pnpm daily:forecast -- --pulse- <path> --pulse-markdown <path> to reuse an existing artifact; AGENT_DECISION_STRATEGY=provider-runtime pnpm daily:forecast for a legacy-engine re-review. After each run, verify the reported env, execute/recommend-only status, runId, and report paths.
How does this skill compare with similar options?
The README contrasts pulse-direct (current default; no external LLM process, parses Pulse structure directly) with provider-runtime (legacy; spawns a Codex/Claude/OpenClaw CLI process). This skill defaults to the former and can switch via AGENT_DECISION_STRATEGY=provider-runtime.