Crowdcore external router

Crowdcore External Router

Use this first when the user starts from a broad product goal and has not selected the exact Crowdcore path.

This skill is a routing and handoff layer. It does not own backend Search OS semantics. It chooses the path, checks boundary configuration, preserves product context, and produces standard artifacts other agent skills can consume.

Core Rules

1. Reply in the user's latest language.
2. Ask at most one blocking clarification question.
3. Before asking, look for local context files: crowdcore-context.json, .crowdcore/context.json, search-context.json.
4. Audience is optional search context. Do not reject public creator search only because no audience is present.
5. Product or brand context must be preserved when available.
6. Do not run live providers in local acceptance. mock and replay are deterministic; live fails closed unless a product API client is explicitly wired.
7. nextActions are proposed actions. If requiresUserConfirmation=true, do not execute the action until the user asks for it.
8. Real Crowdcore API calls require CROWDCORE_API_KEY. If it is missing, tell the user to create an API key in the Crowdcore dashboard API page and set it in the agent environment.

Public Search OS

Use when the user wants to search creators, find influencers, refine creator requirements, or read/export search results.

Minimum executable local acceptance fields:

• platform
• search target or goal
• product or brand context

If enough context exists, run:

node <skill-dir>/scripts/crowdcore-search-os.cjs --mode "${CROWDCORE_SEARCH_AGENT_MODE:-mock}" --context <context-file> --json-out search-results.json --csv-out search-results.csv

After success, report:

• providerMode
• result count
• artifact path: search-results.json
• that Excel export is available after user confirmation

If the user then says 存成 Excel, 导出 Excel, save as Excel, or similar, use the agent's spreadsheet/export capability on search-results.json. Do not rerun search unless the user asks for a new search.

Live Crowdcore Credentials

Mock and replay flows do not require credentials. Live product API or MCP flows require:

export CROWDCORE_API_KEY=cc_live_...
export CROWDCORE_API_BASE=https://api.crowdcore.com

If the user has not created a key yet, direct them to Crowdcore -> API keys -> Create key. The key is shown once and must not be committed to source control.

Other Routes

Private Search: use only when the source pool is explicit, such as list ids or creator ids.

Audience Simulation: use when the request is about ICP synthesis, persona generation, or persistent audience updates.

Ad / Creative / Trend Research: use when the request is about ads, creatives, content patterns, or competitors rather than creator retrieval.

Standard Result Contract

The standard search artifact is crowdcore.search_results.v1. The schema lives at contracts/search-results.schema.json.

Required acceptance fields:

• providerMode
• results
• nextActions
• liveProviderEvaluated=false for mock/replay
• semanticQualityEvaluated=false for mock/replay

The standard export action is:

{
  "type": "export_excel",
  "label": "Export to Excel",
  "requiresUserConfirmation": true
}