菜单
Address suggestions in a past conversation's insight by performing each suggestion's actionable fix. Use when the user asks to implement, apply, address, or act on the suggestions from a conversation insight. Requires a conversation ID. Do NOT use this to generate a new insight.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Create, update, or apply a Holistics dashboard theme (library/themes/) from any input: a brand name, website, screenshot, brand guidelines, explicit colors, or a mood description. Use when the user wants to theme, brand, restyle, or change a dashboard's look and feel. Typical phrasings: build a theme, match our brand, use these colors, make it dark, make it warm/minimal/premium/editorial, glow up this dashboard. Do NOT trigger for building a dashboard, authoring block content, or adding controls.
Create or refine a Holistics custom chart (CustomChartDef in library/custom_charts/) for a chart type the built-in viz can't cover. Use when the user wants to create a custom chart, convert a Vega/Vega-Lite spec into Holistics AML, or scaffold or refine a chart template — or when a dashboard needs a chart form with no built-in equivalent (heatmap matrix, sankey, waterfall, box plot, ...). Typical phrasings: create a custom chart for X, turn this Vega-Lite spec into a Holistics chart, we need a sankey of X. Do NOT trigger for a standard built-in chart or a dynamic content block.
Create, update, or apply a Holistics dashboard theme (library/themes/) from any input: a brand name, website, screenshot, brand guidelines, explicit colors, or a mood description. Use when the user wants to theme, brand, restyle, or change a dashboard's look and feel. Typical phrasings: build a theme, match our brand, use these colors, make it dark, make it warm/minimal/premium/editorial, glow up this dashboard. Do NOT trigger for building a dashboard, authoring block content, or adding controls.
Review a past conversation between a Holistics data-analyst AI agent and a user, then produce insight report with three sections (`Summary`, `Recap`, and `Suggestions`). Use this whenever the user wants to assess, grade, debrief, or critique how the agent performed on a chat, including phrasings like " this conversation", "how did the agent do", "grade this chat", "what could the AI have done better", or "what context should I add to improve the agent", even if they don't say the word "review".
Use this whenever working in a (Holistics) AMQL project (i.e. with AML files). It guides you to set up and review the AMQL development environment and tooling so that the development can be done accurately and efficiently. Tooling includes CLI, MCP, code synchronization, etc.
Search Holistics documentation to answer questions about features, syntax, or concepts. Use this whenever the user asks how something works in Holistics, needs documentation references, or the answer requires looking up Holistics knowledge rather than relying on assumptions.
| name | apply-chat-suggestions |
| description | Address suggestions in a past conversation's insight by performing each suggestion's actionable fix. Use when the user asks to implement, apply, address, or act on the suggestions from a conversation insight. Requires a conversation ID. Do NOT use this to generate a new insight. |
conversation_id (required) — ID of the conversation whose insight suggestions to address. If missing, ask the user.The insight returned by fetch_conversation has this structure:
{
"summary": "Descriptive information about the conversation.",
"recap": "Useful information extracted from the interaction between user and agent.",
"suggestions": [
{ "type": "...", "description": "...", "actionable": "..." }
]
}
suggestions[].actionable items are things to do.summary and recap are background — use them to understand why a fix is needed and how to apply it well, never as fixes themselves.Call fetch_conversation with the given ID. On error (not found, no access, etc.), surface the error and stop.
No insight present → tell the user, ask whether they want to generate one instead, and stop.
Insight present but suggestions empty → tell the user there's nothing to fix. Surface summary / recap for context, and stop.
Suggestions present → STOP. Before calling any TOOL, list the target suggestions so the user sees the full scope This will help users to quickly see which suggestions there are. The suggestions should be a list in this form:
[type] — description
Process each suggestion. Classify the fix and act accordingly:
All changes apply to the current conversation's context, not the original conversation's.
Report results: which suggestions were addressed, which were skipped, and why.