Prompts for designing, operating, and auditing trading agents: decision-loop contracts, risk-envelope reasoning, killswitch pre-flight, first-trade dry runs, and a prompt-injection self-test drawn from the drill corpus. MIT licensed, browsable page by page.
§1Capabilities
Encodes the decision contract an LLM-orchestrated agent runs each cycle, including the non-negotiables: killswitch triggered means hold, envelope breach means hold, ambiguous signals mean hold.
Covers the operational lifecycle: pre-deployment checklists, killswitch verification before every session, structured daily summaries, and a first-trade dry run that reasons through a trade without placing it.
Includes D1-LITE, an open sample of the adversarial drill corpus: ten injection payloads run against your agent’s own decision prompt, with a pass/warn/fail report.
§2Design
Contracts, not vibes. Every prompt specifies inputs, an output schema, and refusal conditions. The agent’s discretion is bounded by design.
Venue data injected. Risk and dry-run prompts expect fees, funding grids and caps as live inputs, never assumed. The Quirks Registry supplies them.
MIT licensed. Use them, modify them, ship them commercially. The open tier is permanently free.
§3Availability by tier
| Open | The open tier, MIT, browsable in full at /prompts. |
| Pro | Monthly updates, venue-tuned variants wired to the Quirks feed. |
| Institutional | The full D1–D7 sentinel corpus with custom drills. |
Indicative until launch.
§4 · Stated plainly
Prompts are templates that bound agent behavior; they are not strategies and not advice. D1-LITE is a sample; the full drill corpus is the Institutional harness.