mnote/rust/crates/mnote-web/src/hermes_tools/doc.rs

use crate::app::AppState;
use crate::context::RequestContext;
use crate::error::WebError;
use crate::hermes_tools::ToolCallInput;
use crate::routes::web_shell::build_page_aggregate_snapshot;
use serde_json::{json, Value};
use std::collections::HashSet;

pub async fn doc_fetch(
    state: &AppState,
    context: &RequestContext,
    input: &ToolCallInput,
) -> Result<Value, WebError> {
    let aggregate = aggregate_value(state, context, input).await?;
    let document_id = input.effective_document_id().unwrap_or_default();
    let workspace_id = input.effective_workspace_id();
    let scope = input
        .arg_string("scope")
        .unwrap_or_else(|| "full".into())
        .to_ascii_lowercase();
    let detail = input
        .arg_string("detail")
        .unwrap_or_else(|| "with_ids".into())
        .to_ascii_lowercase();
    let max_blocks = input
        .arg_value("maxBlocks")
        .and_then(|value| value.as_u64())
        .unwrap_or(120)
        .clamp(1, 240) as usize;
    let mut blocks = block_projection_blocks(&aggregate);
    blocks = match scope.as_str() {
        "outline" => blocks
            .into_iter()
            .filter(|block| block.get("type").and_then(Value::as_str) == Some("heading"))
            .collect(),
        "block" => {
            let block_id = input.arg_string("blockId").ok_or_else(|| {
                WebError::bad_request_code(
                    "mnote_tool_bad_request",
                    "mnote.doc.fetch scope=block 缺少 blockId",
                )
                .with_context(context)
            })?;
            blocks
                .into_iter()
                .filter(|block| block_id_of(block).as_deref() == Some(block_id.as_str()))
                .collect()
        }
        "keyword" => {
            let query = input.arg_string("query").ok_or_else(|| {
                WebError::bad_request_code(
                    "mnote_tool_bad_request",
                    "mnote.doc.fetch scope=keyword 缺少 query",
                )
                .with_context(context)
            })?;
            filter_blocks_by_query(blocks, &query)
        }
        "selection" => {
            let selected_ids = selected_block_ids(input);
            if selected_ids.is_empty() {
                return Err(WebError::bad_request_code(
                    "mnote_tool_bad_request",
                    "mnote.doc.fetch scope=selection 缺少 selectedBlockIds",
                )
                .with_context(context));
            }
            blocks
                .into_iter()
                .filter(|block| {
                    block_id_of(block)
                        .map(|block_id| selected_ids.contains(&block_id))
                        .unwrap_or(false)
                })
                .collect()
        }
        _ => blocks,
    };
    let truncated = blocks.len() > max_blocks;
    blocks.truncate(max_blocks);
    let format = input
        .arg_string("format")
        .unwrap_or_else(|| "json".into())
        .to_ascii_lowercase();
    let include_ids = detail == "with_ids" || detail == "full";
    let content = blocks_to_content(&format, &blocks, include_ids, &document_id, &aggregate);
    let warnings = if truncated {
        json!([{
            "code": "mnote_doc_fetch_truncated",
            "message": "结果已按 maxBlocks 裁剪",
            "maxBlocks": max_blocks
        }])
    } else {
        json!([])
    };
    Ok(json!({
        "ok": true,
        "schema": "mnote.page_ai_context.v1",
        "documentId": document_id,
        "workspaceId": workspace_id,
        "revision": aggregate.pointer("/body/revision").cloned().unwrap_or(Value::Null),
        "conflictDetectionKey": aggregate.pointer("/body/conflictDetectionKey").cloned().unwrap_or(Value::Null),
        "format": format,
        "detail": detail,
        "scope": scope,
        "content": content,
        "blocks": blocks,
        "allowedTargetBlockIds": selected_block_ids(input),
        "truncated": truncated,
        "continuation": if truncated { json!({"maxBlocks": max_blocks}) } else { Value::Null },
        "warnings": warnings
    }))
}

pub async fn doc_find(
    state: &AppState,
    context: &RequestContext,
    input: &ToolCallInput,
) -> Result<Value, WebError> {
    let aggregate = aggregate_value(state, context, input).await?;
    let query = input.arg_string("query").ok_or_else(|| {
        WebError::bad_request_code("mnote_tool_bad_request", "mnote.doc.find 缺少 query")
            .with_context(context)
    })?;
    let match_kind = input
        .arg_string("match")
        .unwrap_or_else(|| "text".into())
        .to_ascii_lowercase();
    let limit = input
        .arg_value("limit")
        .and_then(|value| value.as_u64())
        .unwrap_or(20)
        .clamp(1, 50) as usize;
    let mut matches = Vec::new();
    for block in block_projection_blocks(&aggregate) {
        let matched = match match_kind.as_str() {
            "type" => block
                .get("type")
                .and_then(Value::as_str)
                .map(|value| value.eq_ignore_ascii_case(&query))
                .unwrap_or(false),
            "block_id" | "blockid" => block_id_of(&block).as_deref() == Some(query.as_str()),
            _ => block
                .get("text")
                .and_then(Value::as_str)
                .map(|text| text.contains(&query))
                .unwrap_or(false),
        };
        if matched {
            matches.push(json!({
                "blockId": block.get("blockId").cloned().unwrap_or(Value::Null),
                "type": block.get("type").cloned().unwrap_or(Value::Null),
                "text": block.get("text").cloned().unwrap_or(Value::Null),
                "path": block.get("path").cloned().unwrap_or(Value::Null),
                "revisionRef": block.get("revisionRef").cloned().unwrap_or(Value::Null),
                "score": 1.0
            }));
            if matches.len() >= limit {
                break;
            }
        }
    }
    Ok(json!({
        "ok": true,
        "documentId": input.effective_document_id(),
        "workspaceId": input.effective_workspace_id(),
        "revision": aggregate.pointer("/body/revision").cloned().unwrap_or(Value::Null),
        "conflictDetectionKey": aggregate.pointer("/body/conflictDetectionKey").cloned().unwrap_or(Value::Null),
        "matches": matches
    }))
}

pub async fn plan_update(
    state: &AppState,
    context: &RequestContext,
    input: &ToolCallInput,
) -> Result<Value, WebError> {
    if !input.has_idempotency_key() {
        return Err(WebError::bad_request_code(
            "mnote_tool_idempotency_required",
            "写入计划型 mnote Hermes tool 必须携带 idempotencyKey",
        )
        .with_context(context));
    }
    if input.dry_run != Some(true) {
        return Err(WebError::bad_request_code(
            "mnote_tool_dry_run_required",
            "mnote.doc.plan_update 第一阶段只允许 dryRun=true",
        )
        .with_context(context));
    }
    let aggregate = aggregate_value(state, context, input).await?;
    let command = input
        .arg_string("command")
        .unwrap_or_else(|| "block_replace".into())
        .to_ascii_lowercase();
    let blocks = block_projection_blocks(&aggregate);
    let diff = match command.as_str() {
        "block_replace" => {
            let block_id = required_arg(input, context, "blockId")?;
            let block = find_block(&blocks, &block_id).ok_or_else(|| block_not_found(context))?;
            vec![json!({
                "op": "replace",
                "targetBlockId": block_id,
                "before": block.get("text").cloned().unwrap_or(Value::Null),
                "after": input.arg_value("content").unwrap_or(Value::Null)
            })]
        }
        "block_insert_after" => {
            let anchor = input
                .arg_string("anchorBlockId")
                .or_else(|| input.arg_string("afterBlockId"))
                .ok_or_else(|| {
                    WebError::bad_request_code(
                        "mnote_tool_bad_request",
                        "mnote.doc.plan_update block_insert_after 缺少 anchorBlockId",
                    )
                    .with_context(context)
                })?;
            let block = find_block(&blocks, &anchor).ok_or_else(|| block_not_found(context))?;
            vec![json!({
                "op": "insert_after",
                "anchorBlockId": anchor,
                "after": block.get("text").cloned().unwrap_or(Value::Null),
                "content": input.arg_value("content").unwrap_or(Value::Null)
            })]
        }
        "block_move_after" => {
            let block_id = required_arg(input, context, "blockId")?;
            let anchor = required_arg(input, context, "anchorBlockId")?;
            let block = find_block(&blocks, &block_id).ok_or_else(|| block_not_found(context))?;
            let anchor_block =
                find_block(&blocks, &anchor).ok_or_else(|| block_not_found(context))?;
            let blocked = block_move_after_blocked(&block, &anchor_block, &block_id, &anchor);
            vec![json!({
                "op": "move_after",
                "blockId": block_id,
                "anchorBlockId": anchor,
                "supportedForWrite": !blocked,
                "blocked": blocked
            })]
        }
        "block_delete" => {
            let block_id = required_arg(input, context, "blockId")?;
            let block = find_block(&blocks, &block_id).ok_or_else(|| block_not_found(context))?;
            let blocked = block
                .get("children")
                .and_then(Value::as_array)
                .map(|children| !children.is_empty())
                .unwrap_or(false);
            vec![json!({
                "op": "delete",
                "blockId": block_id,
                "before": block.get("text").cloned().unwrap_or(Value::Null),
                "supportedForWrite": !blocked,
                "blocked": blocked
            })]
        }
        "str_replace" => vec![json!({
            "op": "str_replace",
            "query": input.arg_value("query").unwrap_or(Value::Null),
            "replacement": input.arg_value("content").unwrap_or(Value::Null)
        })],
        other => {
            return Err(WebError::bad_request_code(
                "mnote_tool_bad_request",
                format!("mnote.doc.plan_update 不支持 command={other}"),
            )
            .with_context(context));
        }
    };
    let plan_blocked = matches!(command.as_str(), "block_move_after" | "block_delete")
        && diff
            .first()
            .and_then(|item| item.get("blocked"))
            .and_then(Value::as_bool)
            .unwrap_or(false);
    Ok(json!({
        "ok": true,
        "dryRun": true,
        "planId": format!("plan_{}", context.trace.request_id),
        "documentId": input.effective_document_id(),
        "workspaceId": input.effective_workspace_id(),
        "revision": aggregate.pointer("/body/revision").cloned().unwrap_or(Value::Null),
        "conflictDetectionKey": aggregate.pointer("/body/conflictDetectionKey").cloned().unwrap_or(Value::Null),
        "command": command,
        "diff": diff,
        "warnings": if plan_blocked {
            json!([{
                "code": "block_move_after_blocked",
                "message": "第一阶段仅开放同父级普通叶子块移动，且不能移动到自身之后"
            }])
        } else {
            json!([])
        },
        "risk": if command == "block_move_after" { "medium" } else { "low" },
        "blocked": plan_blocked
    }))
}

fn block_move_after_blocked(
    block: &Value,
    anchor: &Value,
    block_id: &str,
    anchor_id: &str,
) -> bool {
    let same_parent = block.get("parentBlockId") == anchor.get("parentBlockId");
    let leaf = block
        .get("children")
        .and_then(Value::as_array)
        .map(|children| children.is_empty())
        .unwrap_or(true);
    let movable_type = block
        .get("type")
        .and_then(Value::as_str)
        .map(|block_type| matches!(block_type, "paragraph" | "heading" | "todo" | "task"))
        .unwrap_or(false);
    let editable = block
        .get("editable")
        .and_then(Value::as_bool)
        .unwrap_or(false);
    !same_parent || !leaf || !movable_type || !editable || block_id == anchor_id
}

pub(crate) async fn aggregate_value(
    state: &AppState,
    context: &RequestContext,
    input: &ToolCallInput,
) -> Result<Value, WebError> {
    let document_id = input.effective_document_id().ok_or_else(|| {
        WebError::bad_request_code("mnote_tool_bad_request", "页面工具缺少 documentId")
            .with_context(context)
    })?;
    let workspace_id = input.effective_workspace_id();
    let aggregate = build_page_aggregate_snapshot(
        state,
        context,
        &document_id,
        workspace_id.as_deref(),
        None,
        None,
    )
    .await?;
    serde_json::to_value(&aggregate).map_err(|error| WebError::internal(error.to_string()))
}

pub(crate) fn block_projection_blocks(aggregate: &Value) -> Vec<Value> {
    aggregate
        .pointer("/body/blockDocument/blocks")
        .and_then(Value::as_array)
        .cloned()
        .unwrap_or_default()
}

pub(crate) fn block_id_of(block: &Value) -> Option<String> {
    block
        .get("blockId")
        .and_then(Value::as_str)
        .map(str::trim)
        .filter(|value| !value.is_empty())
        .map(ToOwned::to_owned)
}

pub(crate) fn find_block(blocks: &[Value], block_id: &str) -> Option<Value> {
    blocks
        .iter()
        .find(|block| block_id_of(block).as_deref() == Some(block_id))
        .cloned()
}

pub(crate) fn required_arg(
    input: &ToolCallInput,
    context: &RequestContext,
    key: &'static str,
) -> Result<String, WebError> {
    input.arg_string(key).ok_or_else(|| {
        WebError::bad_request_code("mnote_tool_bad_request", format!("工具调用缺少 {key}"))
            .with_context(context)
    })
}

pub(crate) fn block_not_found(context: &RequestContext) -> WebError {
    WebError::bad_request_code("mnote_block_not_found", "块不存在").with_context(context)
}

fn filter_blocks_by_query(blocks: Vec<Value>, query: &str) -> Vec<Value> {
    blocks
        .into_iter()
        .filter(|block| {
            block
                .get("text")
                .and_then(Value::as_str)
                .map(|text| text.contains(query))
                .unwrap_or(false)
        })
        .collect()
}

fn selected_block_ids(input: &ToolCallInput) -> Vec<String> {
    let mut seen = HashSet::new();
    let mut ids = Vec::new();
    for key in ["selectedBlockIds", "allowedTargetBlockIds"] {
        if let Some(Value::Array(values)) = input.arg_value(key) {
            for value in values {
                if let Some(id) = value.as_str().map(str::trim).filter(|id| !id.is_empty()) {
                    if seen.insert(id.to_string()) {
                        ids.push(id.to_string());
                    }
                }
            }
        }
    }
    for key in ["selectedBlockId", "blockId"] {
        if let Some(id) = input.arg_string(key) {
            if seen.insert(id.clone()) {
                ids.push(id);
            }
        }
    }
    ids
}

fn blocks_to_content(
    format: &str,
    blocks: &[Value],
    include_ids: bool,
    document_id: &str,
    aggregate: &Value,
) -> String {
    match format {
        "page_xml" | "xml" => blocks_to_page_xml(blocks, document_id, aggregate),
        "text" | "plain" => blocks_to_text(blocks, include_ids),
        "markdown" | "md" => blocks_to_markdown(blocks, include_ids),
        _ => blocks_to_markdown(blocks, include_ids),
    }
}

fn blocks_to_text(blocks: &[Value], include_ids: bool) -> String {
    blocks
        .iter()
        .map(|block| {
            let text = block_text(block);
            if include_ids {
                format!("[{}] {text}", block_id_of(block).unwrap_or_default())
            } else {
                text
            }
        })
        .collect::<Vec<_>>()
        .join("\n")
}

fn blocks_to_markdown(blocks: &[Value], include_ids: bool) -> String {
    blocks
        .iter()
        .map(|block| {
            let text = block_text(block);
            let prefix = match block.get("type").and_then(Value::as_str) {
                Some("heading") => "## ",
                Some("todo") | Some("task") => "- [ ] ",
                _ => "",
            };
            if include_ids {
                format!(
                    "{prefix}{text} <!-- block:{} -->",
                    block_id_of(block).unwrap_or_default()
                )
            } else {
                format!("{prefix}{text}")
            }
        })
        .collect::<Vec<_>>()
        .join("\n")
}

fn blocks_to_page_xml(blocks: &[Value], document_id: &str, aggregate: &Value) -> String {
    let revision = aggregate
        .pointer("/body/revision")
        .and_then(Value::as_u64)
        .map(|value| value.to_string())
        .unwrap_or_else(|| {
            aggregate
                .pointer("/body/revision")
                .and_then(Value::as_str)
                .unwrap_or_default()
                .to_string()
        });
    let mut output = format!(
        "<page id=\"{}\" revision=\"{}\">",
        escape_xml(document_id),
        escape_xml(&revision)
    );
    for block in blocks {
        let block_id = block_id_of(block).unwrap_or_default();
        let block_type = block
            .get("type")
            .and_then(Value::as_str)
            .unwrap_or("paragraph");
        let revision_ref = block
            .get("revisionRef")
            .and_then(Value::as_str)
            .unwrap_or_default();
        output.push_str(&format!(
            "\n  <block id=\"{}\" type=\"{}\" revisionRef=\"{}\"",
            escape_xml(&block_id),
            escape_xml(block_type),
            escape_xml(revision_ref)
        ));
        if let Some(level) = block
            .pointer("/attrs/level")
            .or_else(|| block.pointer("/props/level"))
        {
            if let Some(level) = level.as_u64() {
                output.push_str(&format!(" level=\"{}\"", level));
            }
        }
        output.push_str(&format!(">{}</block>", escape_xml(&block_text(block))));
    }
    output.push_str("\n</page>");
    output
}

fn block_text(block: &Value) -> String {
    block
        .get("text")
        .and_then(Value::as_str)
        .unwrap_or_default()
        .to_string()
}

fn escape_xml(value: &str) -> String {
    value
        .replace('&', "&amp;")
        .replace('<', "&lt;")
        .replace('>', "&gt;")
        .replace('"', "&quot;")
        .replace('\'', "&apos;")
}