Agent skills for Routescore preflight

routescore-ai is a small library of agent-readable skills. Loaded into a coding agent, they teach it two disciplines:

1. check-swap-preflight — call Routescore's check_swap evidence surface (REST or MCP) before proposing or executing any onchain swap, and read the clear / caution / unsupported verdict as evidence, never as execution approval.
2. verdict-and-caveat-relay — relay that evidence honestly in the final user-facing answer: the verdict as returned, every caveat verbatim, score_state, and the methodology_version — with stale, partial, degraded, or unsupported states represented as gaps, never as confidence.

Routescore is one composable element of an agentic stack — the evidence/attestation element alongside your planner, executor, and wallet elements. The composition the skills assume:

plan → preflight (Routescore) → execute (venue/agent) → record (Routescore)

Install (raw files, no registry)

The skill files are served directly by this site. Fetch them into your agent's skills directory:

BASE=https://www.routescore.io/skills/routescore-ai
mkdir -p .claude/skills/check-swap-preflight .claude/skills/verdict-and-caveat-relay
curl -fsSL "$BASE/skills/check-swap-preflight/SKILL.md" \
  -o .claude/skills/check-swap-preflight/SKILL.md
curl -fsSL "$BASE/skills/verdict-and-caveat-relay/SKILL.md" \
  -o .claude/skills/verdict-and-caveat-relay/SKILL.md

• README.md — the library overview, written for agents to read.
• check-swap-preflight/SKILL.md
• verdict-and-caveat-relay/SKILL.md

Each SKILL.md carries name and description YAML frontmatter — the same grammar other agent skill libraries use — so .claude/skills/ (Claude Code), Cursor rules/context files, and custom frameworks can all load them unchanged. A skill-registry listing (npx skills add …) is on the roadmap, not live today.

What the agent needs

• An API key (free agent tier) for the keyed surfaces the skills call: POST /api/public/v1/check/swap over REST, or the check_swap tool in the @routescore/mcp server — check_swap is on the free agent tier (free 100/day, Pro 1,000, Power 10,000, per account), so any signed-in tier can mint a key and call it. Generate one at Account → Developer → API & MCP access; keys are shown once.
• No key for the browser surfaces the skills fall back to: /route-check and /swap-slippage-checker.

The relay contract, in brief

The skills bind any agent that uses them to four rules:

1. verdict, reasons[], and caveats[] reach the end user verbatim — not paraphrased, softened, or dropped.
2. clear is never a recommendation — it reports the absence of caution-level findings within the stated coverage and score_state.
3. unsupported (HTTP 422) means not evaluated — the agent must not guess, substitute another chain's model, or infer a score.
4. score_state and methodology_version are surfaced with every relayed result.

Every keyed check_swap call also attempts to persist an owner-scoped evidence record and returns its record_id — null, with a record_persistence_failed caveat, if the record store is unavailable. See verifiable pre-sign evidence records for how a third party re-derives the record hash offline.

Check a relay mechanically

The deterministic half of the relay eval harness ships in @routescore/mcp (0.3.0+): given a check_swap response and the agent's final user-facing answer, it reports pass/fail findings per relay rule.

npx -y -p @routescore/mcp routescore-relay-lint response.json answer.txt
# exit 0 = relay contract honored; 1 = fail findings; 2 = usage error