volcengine/searchcli/vs-product-qa
Claude Code and agent skill by volcengine/searchcli
- community
Checking that the SKILL.md file is still in the repository.
Answer Viking AI Search product questions, CLI usage questions, API/auth questions, configuration questions, and troubleshooting questions by grounding every claim in either the installed `vs` CLI's own output or official Volcengine documentation. Never fabricate.
Install
npx vskill@latest install volcengine/searchcli/vs-product-qaSKILL.md
vs-product-qa
When to Use
Use this skill when the user asks grounded questions about Viking AI Search, including:
- product concepts such as scene, dataset, hybrid search, ANN, rerank, or agentic search
- capability or configuration questions such as recommendation setup, boost/bury, or rerank enablement
- API and authentication questions such as AK/SK usage, field semantics, or error meaning
- CLI usage and local troubleshooting such as
vs item apply,vs auth status, or local stack traces - console path, billing, pricing, or quota questions
Do not use this skill for workflow execution. Delegate instead:
- sign-up, purchase, payment, or first AK/SK setup ->
vs-user-onboarding - multi-modal dataset creation and ingestion (records with image/video URLs + text) ->
vs-item-onboarding - web crawling for content ingestion ->
vs-crawler - search tuning suggestions or execution ->
vs-search-tuning
If another skill is active and the user asks a product question outside that skill's scripted scope, answer with vs-product-qa, then return to the original workflow after the answer is complete.
Version Check
Before starting this skill workflow, run vs version check --json. Continue only when status is up-to-date. If status is update-available, stop and tell the user to update the cloned vs repository, then run git pull --ff-only, bash ./scripts/install.sh, and bash ./scripts/install-skills.sh all --target auto --force (PowerShell: scripts/install.ps1 and scripts/install-skills.ps1). If the status is unknown, stop and report that the CLI version could not be verified.
Preconditions
- The installed
vsCLI is available when answering CLI behavior, local auth, LLM configuration, or local error questions. - This skill includes a bundled internal documentation helper under references/volcengine-documentation/SKILL.md.
- Official Volcengine documentation must be fetched through that bundled helper when answering product concepts, API field semantics, purchase, billing, quota, or console UI path questions.
- The agent must classify the question before choosing a source.
- If no selected source can answer, explicitly say the point is not covered by the CLI output or official documentation checked in this turn, then suggest support-ticket / oncall escalation.
Core Principle
Treat the user's installed vs CLI as authoritative for how vs behaves on that machine. For CLI behavior, always check the real CLI first. If CLI output conflicts with documentation, trust the CLI and state that the documentation may be stale.
This skill must not rely on repository source code, generated snapshots, or model memory as the source of truth for customer-facing product answers.
Question Routing
Classify the question before choosing a source.
| Question type | Primary source | Example | |
|---|---|---|---|
| CLI command / flag / usage | installed CLI output: vs <cmd> --help; use the API Reference router only when the user asks how that command maps to an API category | "How do I use vs item apply" | |
| Local authentication / credential | vs auth status / vs doctor | "Why does vs auth status say invalid" | |
| Local CLI error / stack trace | the error's own recovery output | "I see ERR_AUTH_REQUIRED; what now" | |
| Product concept | official docs via the bundled documentation helper | "What is a scene in Viking AI Search" | |
| API contract / request-response shape / field semantics / enum / validation / error code | bundled API Reference, starting from references/api-references/README.md | "What does RelevanceCutoffConfig accept" | |
| Application / dataset effective (valid) data volume | first resolve each bound dataset's Type (via app get / app dataset-config list). For item/video datasets, call app item-data-count per dataset and sum ValidCount (DataNum is not the answer). For document datasets, do NOT call app item-data-count (it returns InvalidParameter: DatasetType) — use app dataset-config list --full (the --full flag is required; the default compact output omits DocumentStats) and read the document count from Config[].Dataset.DocumentStats (DocumentNum + DocumentFromHomepageNum), not DataNum. Always report the document count, including when it is 0. For user_event (behavior) datasets only, exclude them entirely because they have no effective-data-volume metric at all — do not query a count and do not list them (this "omit even when 0" rule applies to user_event only, never to item/video/document, which are always reported even when their count is 0). | "How much effective data does application X have" | |
| Purchase / billing / pricing / quota | official docs via the bundled documentation helper | "How do I upgrade my plan" | |
| Console UI path / where to click | official docs via the bundled documentation helper | "Where do I configure boost/bury" |
Do not use vs --help to answer product-concept questions. Do not use docs to explain command flags when the installed CLI can answer directly.
Knowledge Sources
CLI source
Allowed CLI checks include:
vs --helpvs <cmd> --helpvs doctorvs auth statusvs llm statusvs skill listvs skill search <query>vs skill show <name>
For command-related questions, use vs <domain> --help or vs <cmd> --help as the primary source for command existence, flags, examples, and currently installed behavior. Use the bundled API Reference router only when the user asks how a command maps to an API category or contract.
Before recommending any command or flag, verify that it exists through vs skill list, vs <domain> --help, or vs <cmd> --help.
API Reference source
When the user asks about OpenAPI request/response shape, field semantics, enums, validation rules, or error codes, read the bundled API Reference from references/api-references/README.md.
Read it progressively:
Resolve every relative Markdown link in the API Reference tree against the local directory that contains the current document. For example, ./control-plane/README.md from the root README means references/api-references/control-plane/README.md.
Rules:
- CLI flags and command availability still come from the installed
vs --help; the bundled API Reference must not override observed CLI behavior. - API contract details come from the bundled API Reference.
- If the target API row is marked
missing, say the bundled API Reference is incomplete for that API. Do not guess.
Documentation source
This skill includes a private bundled documentation helper under references/volcengine-documentation/SKILL.md. Use it only as an internal sub-workflow of vs-product-qa. Do not expose it as a separate skill, and do not ask the user to install, trigger, or switch to it.
Use the helper only for official Volcengine documentation, with these fixed rules:
- base root URL:
https://www.volcengine.com/docs/85296/1544972 - Chinese questions use
https://www.volcengine.com/docs/85296/1544972?lang=cn - non-Chinese questions use
https://www.volcengine.com/docs/85296/1544972?lang=en - stay within that root page and its child pages only
- do not access sibling product pages, other documentation roots, site-wide search, homepage navigation, or external sites
- do not switch between
?lang=cnand?lang=enin the same answer unless the user's language changes in a later turn
For Viking AI Search documentation lookup, the data-source restriction is hard:
- product code is
Universal AI Search - the helper script's
searchaction must useServiceCodes="Universal AI Search" - the helper script's
fetchaction must use URLs underhttps://www.volcengine.com/docs/85296only
Documentation lookup protocol:
- Choose the root URL by the user's language.
- If the page URL is already known, or the user provides a page URL under the allowed root, use the helper script's
fetchaction first. - Otherwise, use the helper script's
searchaction withServiceCodes="Universal AI Search". - After identifying the correct page under the same subtree, use
fetchto retrieve the full page content and extract the relevant section. - If exact sub-page lookup fails because of selector change, network error, or timeout, return the selected root URL and tell the user to browse manually. Do not guess sub-page URLs.
Fetch budget:
- at most 3 fetches per answer
- at most 5 seconds per fetch
- total budget 15 seconds or less
- on timeout or overrun, degrade honestly and cite the root URL
Not used in MVP:
https://www.volcengine.com/llms.txthttps://www.volcengine.com/sitemap.xml- undocumented per-page markdown endpoints
These may be used later only if they become publicly available and are actually fetched during the current turn.
Commands
doctor: inspect local CLI environment and service readinessauth status: inspect local Viking AK/SK auth statellm status: inspect local LLM auth state for LLM-backed featuresskill list: list installed skills before recommending handoffskill search: find a relevant installed skill before recommending handoffskill show: inspect another skill's workflow before delegatingvs <cmd> --help: inspect command usage and flags; this is a source rule, not a frontmatter command entry
Workflow
- Classify the question using Question Routing.
- Pick the source:
- CLI usage -> run the relevant
vs ... --help; use the bundled API Reference router only for command-to-API-category questions - local auth / environment -> use
vs auth status,vs doctor, orvs llm status - local CLI error -> use the recovery output from this turn first; only run more CLI checks if needed
- API contract -> read the bundled API Reference from the root README, then follow category and endpoint links
- product docs -> use the bundled documentation helper privately, with the hard
Universal AI Search+docs/85296restriction
- Extract only the lines or sections needed to answer.
- Respond using the required output format.
- If the topic involves credentials, environment variables, or
vs auth import-env, append the AK/SK security notice from references/aksk-notice.md. - If the request implies write operations such as
apply,update,create, orbind, explain only the safe draft or dry-run path unless the user explicitly switches to the correct execution workflow. - If no grounded source can answer, return
unknown, cite the checked CLI command or documentation root URL, and suggest support-ticket / oncall escalation.
Output Format
Use this structure and omit fields that have no grounded content:
CLI entry is only for questions that ask how to use a command. For a data/result question (e.g. "how much effective data does app X have"), omit CLI entry entirely — do not list the commands you ran to obtain the number.
For a data/result question, output only Conclusion. Omit Source, Steps, and CLI entry entirely — do not name the command or backend action, do not add a provenance line, and do not list the commands you ran to obtain the number.
Rules:
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/vs-product-qa/SKILL.md
Popularity
- Repository stars
- 900
- Repository forks
- 24
- 7-day trend score
- 34
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/volcengine/searchcli/vs-product-qa)