SocQ Rednote Research

Collect and analyze public Rednote data

SocQ Links

• Website: https://socq.ai/
• Platform page: https://socq.ai/platforms
• API documentation: https://docs.socq.ai/api-manual/rednote
• API key: https://socq.ai/dashboard/api-key
• MCP and CLI: https://docs.socq.ai/integrations/overview
• Agent Skill guide: https://docs.socq.ai/integrations/skill

Use this Skill to select and run SocQ endpoints for public Rednote research. SocQ requests are asynchronous and credit-metered.

Use When

• The user asks to discover, collect, compare, or analyze public Rednote data.
• The workflow needs Rednote-specific endpoint selection, input validation, pagination, or raw exports.
• An agent must estimate credits, submit a task, poll it to completion, and report normalized results.

Plan Rednote research

Separate note discovery, creator research, and discussion analysis before choosing an endpoint:

• Use rednote/search-notes for topic, product, brand, or keyword discovery. Keep the original query and preserve the returned note URLs for follow-up collection.
• Use rednote/home-recommendations only for a point-in-time recommendation sample. Do not describe it as search results, a stable ranking, or a complete view of the platform.
• Use rednote/note-detail when the user supplies a note URL or when a discovered note needs full metadata before analysis.
• Use rednote/note-comments for top-level discussion and rednote/comment-replies for a selected comment thread. Retain note and parent-comment identifiers so replies remain attributable.
• Use rednote/search-users to resolve an ambiguous creator, then rednote/user-profile for account metadata and rednote/user-notes for that creator's public notes.

For product or brand research, distinguish creator claims, note-level engagement, and audience comments in the final report. Recommendation samples and keyword search results use different discovery mechanisms and must not be merged into a single rank. When comparing creators or notes, align collection windows and requested page limits, and label any partial comment or reply coverage.

Endpoint Selection

• Read platform.md before selecting an endpoint or constructing input. It contains the current endpoint IDs, typed MCP tools, CLI mappings, costs, input choices, and validated examples generated from the Capability Registry.
• Choose from the shape of the requested resource; do not route every request through broad search.
• Prefer direct URLs, canonical usernames, or platform IDs when the user supplies them.
• Re-read the live schema after a validation error instead of inventing parameters.

Key Inputs

• Preserve the user's entities, query terms, date range, locale, filters, ordering, and requested result limit.
• Ask only for missing input required by the selected endpoint.
• Use view: "standard" for MCP and Skill result reads. Use _result_view: "standard" for typed MCP tools or --result-view standard for CLI output.
• Add a reusable idempotency key when a submission might be retried.
• Treat next_cursor as opaque and continue only until the requested scope or user-approved cap is reached.

Execution

1. Prefer an already configured SocQ MCP server at https://api.socq.ai/mcp?platforms=rednote; use the typed tool listed in platform.md, or compact socq_execute when needed.
2. If MCP is unavailable, use socq or npx @socq/cli. Use REST only as the final fallback.
3. Read authentication.md, keep SOCQ_API_KEY in the environment, and never put it in prompts, URLs, committed files, or retained commands.
4. Read billing.md, report the expected cost, and obtain confirmation before a paid large-volume or multi-endpoint run.
5. Submit with _request_source: "skill", --request-source skill, or X-Socq-Source: skill-rest for MCP, CLI, or REST.
6. Save the returned task ID. Treat queued and running as incomplete and follow async-tasks.md until succeeded or failed.
7. Follow pagination.md for all requested pages. Retrieve task files when complete raw JSONL output is required.
8. Read errors.md before retrying authentication, credit, rate-limit, validation, or provider failures.

Output Expectations

Include:

• selected endpoint and execution path (MCP, CLI, or REST)
• concise input and filter summary
• expected and reported credit usage when available
• task ID and terminal status
• result count, pages read, and whether more data remains
• normalized findings or the raw export location
• collection time, failed requests, unsupported filters, and incomplete coverage

Guardrails

• Collect only public data supported by the selected endpoint.
• Do not retry a failed paid request blindly; inspect the normalized error first.
• Do not start a paid large-volume or multi-endpoint run without user confirmation.
• Do not claim completeness when pagination stops early, a provider fails, or a requested filter is unsupported.
• Do not compare metrics collected with different date windows, filters, locales, or content types without labeling the difference.
• Keep task IDs in working notes so interrupted research can resume without resubmitting.