/byted-airesearch-survey

name: byted-airesearch-survey
description: Launch and continue the Volcengine AI Research survey workflow for concept testing, audience design, questionnaire drafting, interview guide generation, execution confirmation, progress checks, and result queries. Use this skill when the user wants to create, revise, confirm, execute, or follow up on a real AI research survey task in ABCompass instead of doing generic brainstorming, copywriting, translation, summarization, or broad market discussion.
license: Apache-2.0
metadata: {"clawdbot":{"emoji":"🧪","homepage":"https://www.volcengine.com/","requires":{"bins":["python3"]},"os":["darwin","linux"]},"openclaw":{"emoji":"🧪","homepage":"https://www.volcengine.com/","requires":{"bins":["python3"]},"os":["darwin","linux"]},"moltbot":{"emoji":"🧪","homepage":"https://www.volcengine.com/","requires":{"bins":["python3"]},"os":["darwin","linux"]}}

Volcengine AI Research Survey

Use this skill to run a real conversational AI Research survey workflow backed by ABCompass.

When to use

Use this skill when the user wants to:

• start a new AI research survey task
• revise an existing research plan
• confirm execution of a generated plan
• check the current plan, progress, or final result
• run concept testing, audience design, questionnaire drafting, or interview-guide generation as a real workflow

Do not use this skill when the user only wants:

• generic brainstorming or ideation with no intention to run a task
• copywriting, translation, polishing, or summarization
• broad market discussion without asking to launch or query the AI Research workflow
• generic platform troubleshooting unrelated to this workflow

Primary command

For every valid workflow turn, run:

python3 scripts/send_survey_message.py --message "<user message>"

This wrapper is the only user-facing entrypoint in this repository. It manages session continuity, API key binding, and production-safe sync transport on top of the canonical `ai_research_message` tool logic.

Core workflow

1. Decide whether the request belongs to the AI Research survey workflow. 2. If the request is valid, call `scripts/send_survey_message.py` before replying. 3. Prefer backend `reply_markdown` as the user-facing answer. 4. Treat backend `presentation`, `next_actions`, `session_event`, and `capability_hints` as the rendering contract. 5. If the task is still running, report real progress only. Do not fabricate a finished report.

Read references only when needed:

• `references/industries.md` for industry normalization guidance
• `references/status-response-rules.md` for status-specific handling
• `references/user-facing-messages.md` for wording consistency

Authentication

First-time use requires an API key.

Preferred input methods:

• `--api-key "<api-key>"`
• `BYTED_AI_RESEARCH_SURVEY_API_KEY`

If no API key is available, ask the user to create or view one at:

• `https://console.volcengine.com/datatester/ai-research/audience/list?tab=apikey`

Then ask the user to provide the API key before calling the API. If the wrapper returns `AUTH_REQUIRED`, do not ask the user to repeat the research request. The pending request can continue directly after key binding.

Parameter guidance

| Category | Parameters | When to set | |----------|-----------|-------------| | Always pass | `--message` | Every call | | Pass when user provides | `--api-key`, `--session-id`, `--force-new-session`, `--research-method`, `--language` | User explicitly specifies these values. Note: `--api-key` is not persisted by the wrapper — the host must retain it in conversation context and pass it on every subsequent call, or guide the user to set the `BYTED_AI_RESEARCH_SURVEY_API_KEY` environment variable. | | Pass when host can confidently resolve | `--request-kind`, `--industry-hint`, `--normalized-message` | Host has enough context to classify intent or normalize industry. Read `references/industries.md` for mapping rules. | | Use from `next_actions` | All parameters in `tool_input` | When the previous response included `next_actions`, use the pre-built parameters directly. Do not modify or guess. | | Leave to defaults | `--source-channel`, `--response-mode`, `--app-id`, `--status-only` | Only set when you have a specific reason |

Critical rules

1. Never bypass `scripts/send_survey_message.py` by manually constructing HTTP requests, curl commands, or calling any API endpoint directly. This wrapper is the only interface to the backend — there are no alternative call paths, sub-agents, or other scripts to try. 2. Never expose API keys, request headers, environment-specific headers, environment variable names, curl commands, or internal request traces in user-facing replies. All debugging output is gated behind the wrapper's built-in redaction mechanism. 3. Never mention or distinguish between internal environments (such as "testing environment", "production environment", "pre-production", or any environment-specific terminology) in user-facing replies. Runtime profile differences are opaque implementation details. 4. Use backend `reply_markdown` verbatim as the user-facing answer. Do not compress, rephrase, summarize, or replace it. If `reply_markdown` is empty, use the fallback message from the response — do not invent content. 5. When the response contains `next_actions`, use them as the exclusive set of suggested next steps. Each action includes a pre-built `tool_input` with all necessary parameters. Do not invent actions outside this set. 6. Do not invent tool methods, parameters, or session lifecycle events beyond what the wrapper supports. 7. Do not force unsupported industries into backend execution. If the industry cannot be confidently mapped, return the unsupported-industry response. 8. For confirm, execute, progress, plan, and result queries, always call the wrapper to get the latest backend state. Never answer from memory or cached context. 9. Do not narrate internal tool steps (readi