Read every open prompt. See what Pro adds.

Decision Loop Scaffold

The core decision contract an LLM-orchestrated agent uses each cycle. Decides what action (if any) to take, respecting the substrate's safety primitives.

Tool-Use Planning Template

Plan a sequence of tool calls (price fetches, position queries, oracle reads, etc.) before committing to an action. Reduces the "agent acts then fetches data" failure mode.

Risk Envelope Reasoner

Evaluate whether a proposed action respects the configured risk envelope, including derived constraints (concentration, drawdown headroom, time-of-day if configured).

Multi-Step Deliberation Chain

When a decision is non-trivial (e.g., conflicting signals, unusual market regime, novel position), run an explicit reasoning chain rather than a single-shot decision.

Self-Critique Loop

Before submitting any action, have the agent critique its own reasoning. Catches single-step errors that survive the initial decision but fail under second-pass scrutiny.

Daily Activity Summary

Produce a structured summary of an agent's last 24 hours of activity. Operator review artifact.

Killswitch Pre-flight Verification

Before an agent session begins, verify all killswitch triggers are armed and consistent with the operator's intent.

Pre-Deployment Checklist

Before promoting an agent from `paper` tier to `restricted` tier (or `restricted` to `full`), run through a structured checklist.

First-Trade Dry-Run Reasoner

Before an agent makes its first live trade on a sub-account, dry-run the proposed first action and identify potential issues.

SDK Setup Walkthrough

Guide an operator through their first SDK installation, given their target chain and their operating system. Conversational; meant to be run interactively.

Sub-Account Configuration Helper

Generate a sub-account configuration block based on the operator's stated risk preferences and instrument focus.

Identity Registration Helper

Generate the agent identity registration call with appropriate metadata.

Prompt-Injection Sentinel (Open-Tier Self-Test)

Lightweight self-test for prompt injection vulnerability. The Open-tier version of the D1 drill; runs ten representative injection payloads against the agent's prompt surface and reports the agent's response category for each.

Risk envelope · quirks-wired variant

The A3 reasoner with venue parameters bound live to the Quirks feed — thresholds update when the venue changes, not when you remember to.

Funding-window scheduler

Plans agent activity around funding windows and measured fee tiers; emits a cadence calendar the decision loop obeys.

Multi-venue failover planner

Generates the failover posture when a venue degrades: what to freeze, what to migrate, what to prove before resuming.

Regime-fabrication drill

Feeds the agent a fabricated market regime and grades whether its decision contract detects the inconsistency before acting.

Stale-data trap drill

Silently freezes selected feeds mid-session and grades hold behavior, staleness detection, and telemetry honesty.

Compromised-host drill

Simulates a hostile execution environment — tampered clock, injected env, poisoned config — and grades containment.

The Interview

your agent learns who it works for

Every bad trading system starts with a mismatch between what the operator says they want and what they can live with.

works with: gryps_capabilities

Explain It Like I'm Five

the six ideas that decide everything

You don't need math to run an agent well.

works with: gryps_venue_snapshot

What Am I Missing?

the agent audits its own blind spots

Agents fail silently at their edges.

works with: gryps_capabilities, gryps_venue_snapshot

Kill Your Favorite Idea

three honest ways to refute yourself

An idea you haven't tried to kill isn't a strategy, it's a mood.

works with: gryps_venue_snapshot

The Question Ladder

what your agent should ask itself, on a schedule

A good trading agent is not the one with the best answers; it's the one that reliably asks the questions whose answers change behavior.

works with: gryps_capabilities

Bend the Coin

test any pattern against dumb luck

"It works 65% of the time" is meaningless until you know the base rate.

Source Audition

make new data earn its seat

Every added source feels like an upgrade and is usually a new way to fool yourself.

works with: gryps_capabilities

The Evidence Ledger

write it down before it happens

Memory is a salesman.