FarmDash Trail Marshal

Read-only DeFi orchestration cookbook for OpenClaw agents. Returns a static catalog of named multi-skill workflow recipes that an agent can use to coordinate research, portfolio review, and (when the user separately installs them) execution sub-skills. Trail Marshal itself produces no analysis, holds no keys, and performs no on-chain action — every state-changing step in every recipe is delegated to the user's separately-installed sub-skill, under that sub-skill's own ClawScan-reviewed contract.

Security Posture. Strict read-only. One tool — list_workflows — returning a static JSON catalog. No environment variables. No credential surface. No outbound calls beyond the catalog endpoint. Trail Marshal does not bundle, invoke, or import the user's other skills; it only describes how an agent could compose them. Every state-changing step in every recipe is owned by a separately-installed sub-skill, under that sub-skill's own ClawScan-reviewed contract and user-confirmation gate.

What this skill does

Trail Marshal is a composition cookbook for OpenClaw agents. It returns a catalog of named multi-skill workflow recipes that an agent can follow when the user's goal spans more than one FarmDash skill (research, portfolio review, swap, perps).

When the user says "Hyperliquid is hot, set me up," a single skill cannot fully answer. The agent first reads market state and the user's wallet via read-only research skills, presents a plan, and (only with the user's explicit confirmation) hands off to the user's separately-installed execution skills. Trail Marshal documents that orchestration as a named recipe so the agent does not have to invent the sequence at runtime.

Available tool

list_workflows

Returns the canonical workflow catalog. Each entry includes a name, goal, required tier, the number of explicit user confirmations the recipe requires, and the sub-skills it composes.

Inputs: Optional category (e.g. farming, risk, accounting), optional tier filter.

Returns shape:

{
  "workflows": [
    {
      "id": "farm_hyperliquid",
      "name": "Farm Hyperliquid Points",
      "goal": "Coordinate a multi-skill points-farming session.",
      "category": "farming",
      "requiredTier": "syndicate",
      "confirms": 2,
      "subSkillsUsed": ["trail-intelligence", "wagon-steward", "signal-architect", "futures-strategist"]
    }
  ]
}

The detailed stepGraph for each recipe is delivered at runtime as part of the list_workflows JSON response and is not embedded in this contract document. Agents fetch the catalog once per session and cache it.

Agent posture: Call list_workflows once at conversation start. Present recipes by name when the user describes a goal that matches one of them.

Workflow catalog (high level)

Ten recipes are published in v0.1. Each is read-only metadata — Trail Marshal never executes any step itself; the user's separately-installed sub-skills do, under their own ClawScan-reviewed contracts.

ID	Goal	Tier	User confirmations
research_only	Help the user understand a protocol without committing capital	Scout	0
airdrop_rotation	Compare farming candidates and rotate via the user's separately-installed swap skill	Pioneer	1 per swap
farm_hyperliquid	Coordinate a Hyperliquid points-farming session	Syndicate	up to 2
farm_solana_restaking	Coordinate a Solana liquid-restaking session	Pioneer	1–2
delta_neutral_setup	Coordinate a paired spot + perp position via separately-installed sub-skills	Syndicate	1
funding_capture	Coordinate a funding-rate strategy on the strongest spread	Syndicate	1
protect_portfolio	Risk-first scan of the user's current positions	Pioneer	0 if no action
idle_capital_deploy	Find idle stables and present highest-yield options	Pioneer	1
migrate_chain	Coordinate moving assets from chain A to chain B	Pioneer	1
bounded_autopilot	Document the contract for delegated, capped autopilot loops (active in v1.0)	Syndicate	1

Each recipe's full step graph is returned at runtime by list_workflows. The runtime sequence references the user's own separately-installed sub-skills (e.g. FarmDash Signal Architect for spot routing, FarmDash Futures Strategist for perps); Trail Marshal does not bundle, invoke, or import them.

Permissions

Trail Marshal v0.1 requires only the user's public wallet address for wallet-scoped recipes. It does not request, accept, or transmit any other credentials. It defines no environment variables. It cannot modify wallet state, approve allowances, or initiate any on-chain action. The single GET endpoint returns a static workflow catalog from https://www.farmdash.one/agents/workflows.json.

If a recipe in the catalog requires execution, the user's separately-installed sub-skill performs that work under its own ClawScan-reviewed contract — Trail Marshal stays out of the call path entirely.

Tier model

Tier	Cost	Limits	Capability
Scout	Free	5 req / 24h	Returns the public workflow catalog
Pioneer	$19.99/mo	500 req / day	Adds workflow filtering by tier and category
Syndicate	$199/mo	50k req / day	Reserved for v1.0 orchestrator engine

Composition patterns

The Sense / Decide / Present / Verify loop

Every Trail Marshal recipe respects this 4-phase pattern:

SENSE    → Read-only research and portfolio skills gather state
DECIDE   → The recipe ranks options and presents a plan to the user
PRESENT  → User reviews; the user's separately-installed sub-skill (if any) handles confirmation under its own contract
VERIFY   → Read-only portfolio skill confirms the new state on the next cycle

If a recipe skips SENSE, it is unsafe. If it skips VERIFY, the agent never improves. Trail Marshal recipes always include both.

Confirmation gate

Trail Marshal cannot pre-confirm anything. Each user confirmation in a recipe happens at the moment the user's separate sub-skill presents its quote — not in Trail Marshal's call path. If asked to run a recipe without intermediate review, the agent should refuse and explain the contract.

Tier composition

A recipe's required tier is the maximum of its sub-skill tiers. When a user is one tier below a recipe's requirement, the agent should recognize the gap from requiredTier, offer the lower-tier subset of the recipe (e.g. research-only without the execution leg), and surface an upgrade link if the user wants the full version. Never silently downgrade a recipe without telling the user what is being skipped.

Reasoning guidelines for agents

• Speak in workflows, not tool soup. Quote a recipe's name and goal when presenting a plan.
• Quote the confirms count up front so the user knows how many user-confirmation steps the recipe involves.
• Re-fetch and re-present if more than ~5 minutes pass between catalog read and the user's review.
• No new analysis. Trail Marshal returns recipe metadata; it does not editorialize or override what other skills produce.

Risk warnings the agent should surface

For any recipe whose execution legs are owned by a separately-installed sub-skill, the agent should restate to the user that:

• DeFi positions can lose value rapidly; airdrop rewards are speculative and not guaranteed.
• Smart contracts can be exploited; even high-Trail-Heat protocols carry technical risk.
• Cross-chain transfers are irreversible if the wrong address or chain is selected.
• Slippage, gas, MEV, and routing fees materially affect realized returns.
• Past Trail Heat performance does not guarantee future scores.
• The user retains full responsibility for every on-chain decision they confirm via their own execution skill.

Affiliate disclosure (inherited from sub-skills)

When a recipe surfaces a https://www.farmdash.one/go/{slug} partnership link, the standard FarmDash affiliate-disclosure block must be included in the same message. Trail Marshal does not introduce new affiliate links of its own.

Versioning

v0.1.x (current). Read-only catalog. The agent drives any execution by composing the user's other separately-installed skills. One tool, one GET endpoint, no credentials.

v1.0 (planned). Optional orchestrator engine — plan_strategy, run_workflow, get_workflow_status — under the Syndicate tier. The v0.1 → v1.0 migration is strictly additive; v0.1 catalog entries continue to work.

Disclaimers

Workflow recipes are documentation, not financial advice. Any action performed by separately-installed sub-skills is the user's decision under that sub-skill's own contract. Trail Marshal performs no on-chain action; the user retains full responsibility for every decision they confirm.

---

Install: Copy this file into your OpenClaw workspace, or fetch https://www.farmdash.one/openclaw-skills/farmdash-trail-marshal/SKILL.md.

Companion skills (separately installed by the user): FarmDash Trail Intelligence (research), FarmDash Wagon Steward (read-only portfolio), FarmDash Signal Architect (spot routing), FarmDash Futures Strategist (perps).

Dashboard: https://www.farmdash.one Agent Hub: https://www.farmdash.one/agents MCP Config: https://www.farmdash.one/.well-known/mcp.json