QuotientSAFE

Bankr Skills equip builders with plug-and-play tools to build more powerful agents.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: quotient
description: >
  Prediction-market intelligence for Polymarket agents. Quotient runs a multi-role AI
  forecasting pipeline over 1,600+ sources and publishes daily trade signals with side,
  entry prices, conviction tiers, capacity, and convergence reads. Pull forecasts (with
  what-changed deltas), recent sources (articles + X posts), the featured signal, the
  daily WTI crude oil read, and per-wallet portfolio intelligence; execute via Bankr.
  Pays via x402 in USDC on Base or USDG on Robinhood Chain.
  Triggers on: "quotient signals", "trade signals", "featured signal", "oil signal",
  "WTI", "crude", "what's new with my portfolio", "hold or sell", "convergence",
  "mispriced markets", "what does Q think", "quotient odds", "prediction market
  intelligence", "polymarket intelligence", "recent sources for", "what markets does
  quotient have", "market forecast", "should I bet on".
emoji: 🔮
tags: [polymarket, prediction-markets, trading, intelligence, x402]
version: 1.0.0
visibility: public
metadata:
  clawdbot:
    emoji: "🔮"
    homepage: "https://quotient.social"
    requires:
      bins: ["curl", "jq", "node", "bankr"]
credentials:
  - name: BANKR_API_KEY
    description: Only needed for signal-strategy.mjs --execute (Bankr Agent API key, read-write).
    required: false
    storage: env
---
<!-- GENERATED from public/skill/skill.md — edit there, then npm run skill:build -->

# Quotient API Skill

Quotient = intelligence. Bankr = execution. This skill reads Quotient's x402-paid API for
forecasts, published trade signals, sources, the oil read, and wallet portfolio
intelligence, then hands off to Bankr natural-language prompts for any trade. Nothing
here places trades directly.

## Base URL & Discovery

- `QUOTIENT_BASE_URL`: `https://quotient-api-gateway.onrender.com`. The scripts enforce an
  exact HTTPS origin allowlist on it — the default gateway origin is hardcoded, and extra
  origins can be added only through the local policy file
  (`references/payments-policy.md`), never via env or fetched content.
- Discovery, same origin: `/openapi.json` (canonical routes + params), `GET /api/public/pricing`
  (billing metadata), `/llms.txt` (AI index), `/skill/*` (these docs + scripts)
- Treat OpenAPI as canonical invocation metadata; treat the runtime `402` challenge as the
  authoritative price.

## How Q Works

Quotient's forecasting agent (Q) runs a multi-role analysis pipeline on every market it
covers: question analysis, research, base-rate analysis, bull/bear advocacy, contrarian
examination, and synthesis, pulling from 1,600+ sources. Each run produces an independent
probability estimate, a BLUF (bottom-line-up-front) thesis, key drivers with citations,
and delta-from-prior reasoning. A separate publisher watches for markets where Q diverges
materially from the venue price and publishes a small number of trade signals per day.

See https://quotient.social for the current live track record.

Coverage is strongest on world-events markets — Iran, tariffs, elections, central-bank
policy, conflict escalation, diplomatic negotiations. If it moves geopolitical risk, Q
probably has a view.

## Key Concepts

**Markets** — Prediction markets Quotient covers. Each has a `slug` (Polymarket slug),
question, current `market_odds`, dispute status, and Q's forecast history.

**Forecasts** — Q's probability estimate for a market, refreshed as new material lands.
Every forecast carries the change primitives: `delta_from_prior` (how much Q moved),
`delta_reasoning` (a deterministic sentence saying why), `refresh_reason` /
`refresh_triggered_by` (non-null = the rerun was triggered, not scheduled), plus
`headline`/`bluf`/`crux` and conviction inputs (`draw_std_log_odds`, `draw_count`,
`band25`/`band75`). "What changed" is read straight off the node — never inferred.

**Trade signals** — Published `:QuotientSignal` entries: Q's actual calls, a handful per
day. A signal can remain active for up to seven days; the latest forecast can refresh many
times during that hold. Read `published_at`/`is_new_today` for publication context,
`forecast_updated_at`/`is_fresh` for research freshness (six-hour threshold), and
`is_active` for lifecycle state. The `/signals` `window` filters forecast updates, not
publication time, and its default feed omits `paused`, `done`, and `retired`
rows. It returns at most one signal per market: the newest publication is selected before
side/status/conviction filters, with no fallback to an older signal when that newest call is
ineligible. Each signal also has a `side` (YES/NO), entry prices (`entry_q` = Q at publish,
`entry_pm` = market at publish, `entry_spread_pp` = the gap in points), a board `status`, a
conviction tier, capacity, and a live-priced convergence read:

- `status`: `actionable` (buyable now) · `unconfirmed` (Q's latest forecast flipped side
  vs prior — wait for confirmation) · `paused` (temporarily unavailable after a deep drawdown, venue divergence, or safety veto — do not
  chase) · `done` (converged: `converge_upside_pct ≤ 0`, thesis played out) · `retired`
  (terminal; `retired_reason` ∈ `resolved` | `flipped` | `fading_q` | `expired`).
- Conviction: `conviction_tier` 1–3 from forecast-ensemble dispersion (3 = tightest),
  mirrored as `conviction` high/medium/low; `has_band` is `false` only when no conviction
  read could be computed at all (missing Q or price) — pre-ensemble inferred reads still
  report `true` with tier capped at 2.
- Convergence (all cents on Q's side of the book): `q_value_cents` (Q's value),
  `entry_cost_cents`, `current_cost_cents`, `distance_to_convergence_cents`,
  `converge_upside_pct`. `live_priced` + `priced_at` disclose whether the read used a
  live CLOB midpoint or a graph fallback.
- Capacity: `capacity_usd_at_2c` (near-touch depth), `capacity_basis`
  (`depth-2c` | `volume-fallback` | null), `capacity_available`, `capacity_as_of`.

**Pre-trade liquidity report (required before any buy handoff):** tell the us

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

declared (2 observation(s))

Network

declared (9 observation(s))

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (2)

|| "$base" == http://127.0.0.1:* || "$base" == http://127.0.0.1 ]]; then

Every endpoint in this file is a public read — no API key, no wallet signature, no spend. Use

Gates applied: no_behavioural_pass.