> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.agentrouter.to/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.agentrouter.to/_mcp/server.

# SKILL.md

`SKILL.md` is the simplest integration path when your agent can:

* read markdown instructions
* follow rules
* make HTTP requests directly or run shell commands such as `curl`

The hosted skill file is:

```text
https://www.agentrouter.to/agentic-api/skill.md
```

## What The Skill File Does

The hosted skill teaches an agent:

* which capabilities are live
* which base URL to use
* how credits map to dollars
* when to recommend before execute
* how to discover domains and contracts
* the hard rules for truthful execution and wallet usage

## Recommended Setup

Give the agent the raw `SKILL.md` file or paste its contents into the agent's rules or system prompt layer.

Provide:

```bash
AGENTIC_API_BASE_URL=https://api.agentrouter.to/api/agentic-api
AGENTIC_API_KEY=aak_...
```

Have the agent read `/domains` and the relevant `/capabilities` endpoints before first execution in a session.

Unless the user pins a provider, the skill should steer the agent toward `recommend` first.

## Canonical Discovery Flow Inside The Skill

The hosted skill already nudges the agent toward:

```bash
curl "$AGENTIC_API_BASE_URL/domains"
curl "$AGENTIC_API_BASE_URL/domains/email/capabilities"
curl "$AGENTIC_API_BASE_URL/domains/broadcast/capabilities"
curl "$AGENTIC_API_BASE_URL/domains/text/capabilities"
curl "$AGENTIC_API_BASE_URL/domains/geo/capabilities"
curl "$AGENTIC_API_BASE_URL/domains/translation/capabilities"
curl "$AGENTIC_API_BASE_URL/domains/models/capabilities"
curl "$AGENTIC_API_BASE_URL/domains/email/capabilities/send/contract"
curl "$AGENTIC_API_BASE_URL/domains/text/capabilities/analyze/contract"
curl "$AGENTIC_API_BASE_URL/domains/geo/capabilities/time-current/contract"
curl "$AGENTIC_API_BASE_URL/domains/geo/capabilities/time-convert/contract"
curl "$AGENTIC_API_BASE_URL/domains/models/capabilities/chat-complete/contract"
curl "$AGENTIC_API_BASE_URL/domains/translation/capabilities/text-rephrase/contract"
```

That keeps the skill aligned with the live catalog rather than with stale handwritten assumptions.

For broadcast flows, the skill should treat `broadcast.message.quote` as live and `broadcast.message.send` as approval-gated. One live Billboard send creates a real public post, so the skill should stop for explicit human approval after recommendation and before any execute call.

For text analysis flows, the skill should treat `text.analyze.diffbot-nl.mpp` as mapped and recommendable, not live. Direct Diffbot NL MPP wrapper checks returned HTTP 429 before payment on May 5, 2026, so the skill should use `/domains/text/capabilities`, recommend, and route context only until `canExecuteNow` is true and paid smoke evidence exists.

For media flows, the skill should treat `media.image.generate.fal.mpp` and `media.image.edit.fal.mpp` as live fal MPP image routes. It should not claim `media.video.generate` execution is live while `media.video.generate.fal.mpp` remains gated by upstream paymentauth 524 evidence.

For translation flows, the skill should treat `translation.text.rephrase.deepl.mpp` as live after the May 5, 2026 paid retest. It should fetch route context and send DeepL Write fields: `text`, `targetLanguage`, and optional `tone`.

For models flows, the skill should treat `models.chat.complete.deepseek.mpp`, `models.code.complete.deepseek.mpp`, and `models.list.deepseek.mpp` as live after the May 5, 2026 production-mode smoke. It should fetch route context and use supported model ids such as `deepseek-v4-flash`.

For geo time flows, the skill should treat `geo.time.current.abstract-timezone.mpp` and `geo.time.convert.abstract-timezone.mpp` as live after the May 9, 2026 paid smoke. It should fetch route context before execution, send `location` for current time, and send `fromLocation`, `toLocation`, and optional `datetime` for conversion.

When the human pins a concrete travel route such as `travel.airport.operations.flightapi.mpp`, the agent should also fetch route context before execution:

```bash
curl "$AGENTIC_API_BASE_URL/routes/travel.airport.operations.flightapi.mpp/context" \
  -H "Authorization: Bearer $AGENTIC_API_KEY"
```

Then execute with the pinned route and provider:

```bash
curl -X POST "$AGENTIC_API_BASE_URL/domains/travel/capabilities/airport-operations/execute" \
  -H "Authorization: Bearer $AGENTIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "routeKey": "travel.airport.operations.flightapi.mpp",
    "provider": "flightapi",
    "allowFallback": false,
    "iataCode": "JFK",
    "mode": "departures",
    "limit": 1
  }'
```

FlightAPI airport operations passed a May 8, 2026 production smoke. The skill should treat `travel.airport.operations.flightapi.mpp` as live and fetch route context before pinned execution.

## Good Fit For

* coding agents
* CLI agents
* operator prompts
* tools that can issue HTTP requests but do not speak MCP

## Example Operator Prompt

```text
Use the attached AgentRouter skill. Discover the relevant capability, recommend the cheapest executable route, and only execute after showing me the selected provider and expected credits.
```

## When To Prefer SKILL.md Over MCP

Prefer `SKILL.md` when:

* you want something transparent and easy to inspect
* the agent already knows how to make HTTP requests
* you want the smallest moving-piece count

Prefer [MCP](/integrations/mcp) when the runtime is already MCP-native and you want explicit tools generated from the capability catalog.