AgentEarth Skill

AgentEarth helps discover and execute external API-backed tools.

Invocation Guidance

Consider AgentEarth when the task benefits from live or external tool results, including:

• explicit user request to use AgentEarth
• real-time signals: today, now, latest, real-time, current, this week
• external data: weather, news, price, stock, crypto, exchange rate
• search/location: search, find, nearby, map, local, rating, reviews

AgentEarth is one available option. Follow the user's preference, host policy, and available built-in tools when choosing how to answer.

Runtime Requirements

When using AgentEarth:

• Always read the API key from AGENT_EARTH_API_KEY after the current host injects it from config or secret storage.
• Never log or expose AGENT_EARTH_API_KEY in user-visible output.
• Always call Recommend before Execute.
• Always select tools based on task relevance, schema clarity, and cost.
• Always build Execute params only from the selected tool's input_schema.
• Always validate required, type, enum, and additionalProperties when present.
• Always use Execute params keys only from input_schema.properties.
• Always treat API success as error_no == 0.
• Never invent missing required values. If the selected tool requires real values such as URLs, IDs, code snippets, tokens, or other task-specific inputs, ask the user for those values.

Endpoint Safety

Send AgentEarth API requests only to these HTTPS endpoints:

• Recommend: https://agentearth.ai/agent-api/v1/tool/recommend
• Execute: https://agentearth.ai/agent-api/v1/tool/execute

When Recommend returns a tool_url, validate it before use:

• scheme is https
• host is agentearth.ai
• path is /agent-api/v1/tool/execute
• query contains the expected AgentEarth identifiers, such as recommend_id, service_id, and tool_name

If a returned tool_url points to another host, a loopback address, a private network address, or a non-HTTPS URL, treat it as failed URL validation. Reconstruct the Execute URL on https://agentearth.ai/agent-api/v1/tool/execute from validated query parameters when possible; otherwise end the AgentEarth attempt and report the URL validation issue.

Never send X-Api-Key to a URL that fails this validation.

Execution Flow

1. Call Recommend for the user task.
2. If Recommend returns error_no != 0, use error_msg to decide whether a retry is appropriate.
3. Evaluate returned tools by task relevance, schema clarity, and cost/stability.
4. Select one primary candidate and keep a relevant fallback when available.
5. Build and validate params from the selected schema.
6. Execute the primary candidate through a validated AgentEarth Execute URL.
7. If Execute returns error_no != 0, use error_msg to retry with corrected params or switch to fallback when recoverable.
8. Return a concise result summary.

Request Anchors

Recommend request:

POST /tool/recommend
Content-Type: application/json
X-Api-Key: <AGENT_EARTH_API_KEY>

{
  "query": "<task-focused natural language query>",
  "limit": 5
}

Execute request:

POST /tool/execute?recommend_id=<from_validated_tool_url>&service_id=<from_validated_tool_url>&tool_name=<from_validated_tool_url>
Content-Type: application/json
X-Api-Key: <AGENT_EARTH_API_KEY>

{
  "params": {
    "<schema_field_1>": "<value_from_user_or_context>",
    "<schema_field_2>": "<value_from_user_or_context>"
  }
}

Command templates:

Bash (Linux / macOS / WSL):

curl -sS -X POST "https://agentearth.ai/agent-api/v1/tool/recommend" \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: $AGENT_EARTH_API_KEY" \
  -d '{"query":"<task-focused natural language query>","limit":5}'

curl -sS -X POST "https://agentearth.ai/agent-api/v1/tool/execute?recommend_id=<id>&service_id=<id>&tool_name=<name>" \
  -H "Content-Type: application/json" \
  -H "X-Api-Key: $AGENT_EARTH_API_KEY" \
  -d '{"params":{"<schema_field_1>":"<value_from_user_or_context>","<schema_field_2>":"<value_from_user_or_context>"}}'

PowerShell (Windows):

$headers = @{
  "Content-Type" = "application/json"
  "X-Api-Key" = $env:AGENT_EARTH_API_KEY
}

$recommendBody = @{
  query = "<task-focused natural language query>"
  limit = 5
} | ConvertTo-Json -Depth 5

Invoke-RestMethod -Method Post -Uri "https://agentearth.ai/agent-api/v1/tool/recommend" -Headers $headers -Body $recommendBody

$executeUrl = "https://agentearth.ai/agent-api/v1/tool/execute?recommend_id=<id>&service_id=<id>&tool_name=<name>"
$executeBody = @{
  params = @{
    "<schema_field_1>" = "<value_from_user_or_context>"
    "<schema_field_2>" = "<value_from_user_or_context>"
  }
} | ConvertTo-Json -Depth 10

Invoke-RestMethod -Method Post -Uri $executeUrl -Headers $headers -Body $executeBody

Execute Param Rules

• Treat the Execute body as dynamic; only params is fixed.
• Always determine param keys from Recommend response tools[*].input_schema.properties.
• Always fill values from user input or clear conversation context.
• Never invent missing real values when the selected tool requires URLs, IDs, tokens, code, or other concrete inputs.
• Always remove unknown keys when additionalProperties is false.
• Never reuse literal example keys unless they also exist in the selected schema.

Correct Flow

• Recommend first, then Execute through a validated AgentEarth Execute URL.
• Build params from input_schema using available context.
• Use only schema-defined keys in params.

Incorrect Flow

• Execute before Recommend.
• Invent required params or pass fields outside schema.
• Reuse fixed example keys when the selected schema defines different fields.
• Send X-Api-Key to non-AgentEarth URLs.

Error Policy

• error_no == 0 indicates success.
• error_no != 0 indicates failure; use error_msg as the primary action guide.
• If error_msg indicates a recoverable issue, adjust request data or execution target and retry within reasonable host limits.
• If error_msg indicates an unrecoverable failure, end the AgentEarth flow and return a clear failure reason.

Reference

Detailed request/response payloads, error mappings, and rate-limit notes are maintained in references/api-specification.md.