Add ToolSearch registry discovery

Author: SqlRush

docs/cc-100-roadmap.md

@@ -191,6 +191,8 @@ M8 补充：command registry 现在保存本地 skill prompt template，并提
 
 M8 补充：新增基础 `Skill` tool wrapper，已注册到默认内置工具集，可按官方 `skill`/`args` 输入调用项目目录发现到的本地 prompt skill，并兼容 `commandName`/`arguments` 别名；tool result 会返回 `Launching skill: ...`、structured command metadata 和 prompt expansion 生成的 meta user message，conversation runner 现在会把 `ToolResult.NewMessages` 追加进后续模型请求和 transcript。插件目录 skill 可通过工作目录发现和 display name 调用，structured content 会保留 source、loadedFrom、displayName、description、argument metadata、skillRoot、whenToUse、version、context、agent、effort、paths、contentLength 和 progressMessage，方便后续 UI/SDK/remote surface 使用。forked skill、remote/MCP/bundled skill 细节、skill prompt shell injection、slash/local command UI wiring 仍未完成。
 
+M8 补充：新增基础 `ToolSearch` tool 并注册到默认内置工具集，executor 会把当前 tool registry 放入工具 metadata，`ToolSearch` 可按 name、alias、description、prompt、search hint 搜索当前可用工具定义，返回 ranked structured results 和 read-only/concurrency/destructive 元数据，并兼容 `query`/`q`/`search` 与 `topn`/`limit` 等输入别名。完整 deferred/lazy tool discovery 和官方 BM25 等价排序仍未完成。
+
 M8 补充：新增基础 slash command parser/executor，按官方 `/command args` 与 `/mcp:tool (MCP) args` 形态解析，并把本地项目 prompt skill slash 调用接入 conversation runner：`/skill args` 会生成 `<command-name>/<command-message>/<command-args>` metadata user message 和展开后的 meta prompt message，写入 transcript/parent chain 后再请求模型；skill frontmatter `model` 可覆盖本轮请求模型。local/local-jsx 命令目前只返回未实现输出，不会误发给模型；command permissions attachment、forked skill、MCP/plugin/bundled slash 来源和 UI 仍未完成。
 
 M8 补充：本地 prompt skill 的 slash 调用和 `Skill` tool 现在都会生成 `command_permissions` attachment，按官方 `allowed-tools` 解析 comma/space 分隔且保留括号内模式；conversation runner 会在当前 turn 内把这些 `PermissionSourceCommand` allow rules 合并进 engine permission decider，让 skill frontmatter 授权的后续工具调用可在同一轮放行，并继续保留 model override attachment metadata。完整权限 UI 展示、SDK event surface、forked/MCP/plugin/bundled skill 权限继承仍未完成。

docs/claude-code-go-rewrite-plan.md

@@ -290,6 +290,7 @@ test/parity/                 # golden tests against TS/official behavior
 - 本轮补充：新增 `internal/commands` registry 基础层，按官方 command 来源顺序合并 bundled/builtin-plugin/project-skill/workflow/plugin/dynamic/builtin metadata，并提供 dynamic 去重、display-name/alias 查找、hidden 过滤、SkillTool/slash-skill 过滤和 bridge-safe 判定；实际 local/local-jsx 执行、`/help`/`/skills` UI、plugin/MCP/workflow 加载仍未宣称完成。
 - 本轮补充：command registry 现在保存本地 skill prompt template，并提供 prompt expansion 入口，覆盖 `$ARGUMENTS`、indexed/shorthand 参数、frontmatter named arguments、`${CLAUDE_SESSION_ID}` 替换、非 MCP skill 的 `${CLAUDE_SKILL_DIR}` 替换和 meta user message 输出；shell injection、SkillTool wrapper、local/local-jsx 执行和 REPL/UI wiring 仍未宣称完成。
 - 本轮补充：新增基础 `Skill` tool wrapper 并注册到默认内置工具集，可调用本地项目 prompt skill、兼容官方 `skill`/`args` 与相邻别名，并把 prompt expansion 产生的 meta user message 通过 `ToolResult.NewMessages` 交给 conversation runner；runner 现在会把这些新消息写入 transcript 并追加到下一轮模型请求。插件目录 skill 可通过工作目录发现和 display name 调用，Skill tool structured content 会保留 source、loadedFrom、displayName、description、argument metadata、skillRoot、whenToUse、version、context、agent、effort、paths、contentLength 和 progressMessage。forked/remote/MCP/bundled skills、plugin skill UI/SDK polish、shell injection、slash/local command UI wiring 仍未宣称完成。
+- 本轮补充：新增基础 `ToolSearch` tool 并注册到默认内置工具集，executor 会把当前 tool registry 注入工具 metadata；`ToolSearch` 可按 name、alias、description、prompt、search hint 搜索当前工具定义，返回 ranked structured results 和 read-only/concurrency/destructive 元数据，并兼容 query/limit 相邻输入别名。deferred/lazy tool discovery 与官方 BM25 等价排序仍未宣称完成。
 - 本轮补充：新增基础 slash command parser/executor，支持官方 `/command args` 和 `/mcp:tool (MCP) args` 解析；conversation runner 现在会在请求模型前展开本地项目 prompt skill slash command，生成 command metadata user message 和 meta prompt message，保留 transcript parent chain，并支持 skill `model` 覆盖本轮请求。local/local-jsx 目前只返回未实现输出且不会误发模型，command permissions attachment、forked/MCP/plugin/bundled slash command 和 UI wiring 仍未宣称完成。
 - 本轮补充：本地 prompt skill slash command 和 `Skill` tool 现在都会生成 `command_permissions` attachment，解析 `allowed-tools` 的 comma/space 分隔形式并保留括号内 tool pattern；Runner 会在当前 turn 内把 attachment 转成 `PermissionSourceCommand` allow rules 合并进 engine permission decider，让 skill 授权的后续工具调用可在同一轮通过。完整权限 UI/SDK 展示、forked/MCP/plugin/bundled skill 权限继承仍未宣称完成。
 - 本轮补充：skill frontmatter 标量兼容继续补齐，`allowed_tools`/`argument_hint`/`disable_model_invocation`/`user_invocable`/`when-to-use` 等相邻字段会映射到 canonical command metadata；`model: inherit` 不再误触发模型覆盖，`context: fork`、`agent`、`effort` 会保留在 command contract 中，为后续 forked skill/agent 执行接线提供 metadata；当 policy 锁定 `agents` surface 时，非可信来源 prompt command 的这些 agent metadata 会在 registry 层清除，plugin/bundled/admin 来源保留。

docs/first-second-parity-audit.md

@@ -88,6 +88,7 @@ Anthropic API 和 conversation：
 - `/plugin <plugin-name>` now takes the no-query local plugin-detail path used by `/plugin show <plugin-name>` when the name matches a local plugin manifest, while unknown one-word subcommands still report unsupported.
 - `/model show`, `/model info`, and `/model current` now take the no-query current-model path instead of treating those words as custom model names and mutating the runner model.
 - A basic `Skill` tool wrapper is now registered with the default built-in tool set. It can invoke local project and plugin-directory prompt skills through the command registry, returns the official-style `Launching skill: ...` result plus structured command metadata, preserves command source/display/root/frontmatter metadata for plugin skills, and passes expanded meta user messages through `ToolResult.NewMessages`; the conversation runner now appends those new messages to transcripts and subsequent model requests.
+- A basic `ToolSearch` tool is now registered with the default built-in tool set. The executor exposes the current registry through tool metadata, and `ToolSearch` can search available definitions by name, alias, description, prompt, and search hint, returning ranked structured results with read-only/concurrency/destructive metadata and query/limit aliases.
 - Basic slash prompt command invocation is now wired for local project prompt skills: `/skill args` is parsed before the model request, converted into official-shaped command metadata plus expanded meta prompt messages, appended to transcript parent chains, and can override the turn model from skill frontmatter.
 - Local prompt-skill slash commands and the `Skill` tool now emit `command_permissions` attachments, parse `allowed-tools` into command-scoped permission rules, and merge those rules into the engine permission decider for the current turn so authorized follow-up tool calls can pass without leaking into later turns.
 - Settings schema coverage was expanded for major Claude Code settings keys: auth helpers, model allowlists/overrides, MCP policy, hooks policy, worktree, shell, output style, language, thinking/effort, plugins, remote, spinner, sandbox, and related flags. Settings parsing now also coerces `env` values to strings, filters invalid permission rules into warnings, validates key permission fields, applies WebSearch/WebFetch-specific permission-rule validation, validates marketplace source union basics across `extraKnownMarketplaces`, `strictKnownMarketplaces`, and `blockedMarketplaces` including settings-source map key/source name consistency, plugin item path shape, and `installLocation` user/project/local values, derives runtime marketplace allow/block decisions from strict and blocked settings, loads installed plugins from project-chain and user plugin roots with project precedence, loads marketplace-only plugin listings for headless `/plugin available` and `/plugin marketplace plugins|search|show` while preserving installed plugin available/update state, exposes installed/available marketplace JSON through CLI `plugin list --json --available`, exposes configured marketplace list JSON/text through CLI `plugin marketplace list`, writes and removes configured marketplaces to user/project/local settings scopes through CLI `plugin marketplace add/remove --scope ...`, triggers marketplace cache refresh through CLI `plugin marketplace update [name]`, shares marketplace install copy/conflict/symlink-safety code between headless `/plugin install [--scope project|user|local]` and CLI `plugin install --scope project|user|local` with marketplace `installLocation` defaults, shares marketplace update/replace code between headless `/plugin update [--scope project|user|local|all] [name]` and CLI `plugin update --scope project|user|local|all <plugin>` with the same named-plugin defaulting, writes plugin enable/disable state to user/project/local settings scopes through CLI `plugin enable|disable --scope ...`, loads managed policy from macOS MDM plist / Windows HKLM, platform `managed-settings.json`, sorted `managed-settings.d/*.json` drop-ins, optional remote managed-settings GET sources, and Windows HKCU fallback, refreshes remote managed policy at turn start and during daemon heartbeat ticks with merged-settings/plugin-server app-state propagation, honors `allowManagedPermissionRulesOnly` for permission-rule merging, enforces `strictPluginOnlyCustomization` for the implemented `skills`, `mcp`, and `hooks` surfaces so user/project/local customizations are skipped while plugin/admin sources remain, strips untrusted prompt-command `agent`/`context: fork`/`effort` metadata when the `agents` surface is policy-locked while preserving plugin/bundled/admin sources, includes a path-keyed settings file cache plus a snapshot-based settings change detector that classifies created/modified/deleted settings files and resets cached reads on change, refreshes local user/project/local settings at turn start with plugin-server app-state propagation, and now generates a JSON Schema from `contracts.Settings` with `/config show schema` summary visibility.
@@ -106,7 +107,7 @@ The following items remain incomplete and must not be treated as done:
 - Auto mode / YOLO classifier: transcript construction, two-stage classifier, XML/tool-use parsing, prompt dump, denial circuit breaker, model gating, and fallback behavior.
 - Interactive permission prompt flow: REPL dialogs, bridge/channel/swarm permission relays, user feedback images, prompt race handling, cancellation.
 - Full filesystem permission parity gaps that remain: skill-scope allow suggestions, remaining local/local-jsx slash-command execution wiring beyond the currently covered no-query builtins, command permission UI/SDK surfacing, forked/remote/MCP/bundled SkillTool behavior and remaining plugin SkillTool UI/SDK polish beyond current metadata preservation, skill prompt shell injection, complete auto-memory override policy, and deeper platform-specific Windows/WSL bypass handling.
-- Full tool execution parity gaps that remain: complete hook runtime policy beyond the current settings command hook path, MCP elicitation, complete SDK control event surface beyond current progress NDJSON, mid-call cancellation for concrete tools, background task behavior, telemetry, schema-not-sent hints, and concrete tool-specific semantics.
+- Full tool execution parity gaps that remain: complete hook runtime policy beyond the current settings command hook path, MCP elicitation, complete SDK control event surface beyond current progress NDJSON, deferred/lazy tool discovery and official BM25-equivalent tool search beyond the current local registry search, mid-call cancellation for concrete tools, background task behavior, telemetry, schema-not-sent hints, and concrete tool-specific semantics.
 - Complete Anthropic API parity gaps that remain: dynamic beta-header latching for any remaining provider-specific feature gates beyond prompt-cache/cache-editing/strict-output/1M-context requests, ant-only dump gating and remote `/issue` submission integration beyond the current local context summary, full official cost tracker/session restore lifecycle beyond current transcript usage restore, post-handshake streaming failure semantics, deeper gateway/proxy auth and routing behavior beyond static custom headers, first-party/Bedrock/Vertex/Foundry client setup, deeper OAuth account/profile lifecycle beyond request-time refresh retry, fast-mode retry/cooldown semantics, persistent unattended retry heartbeats, full prompt-cache editing lifecycle, and provider-specific cache behavior.
 - Full conversation/query loop: stop hooks, compact/auto-compact, token budget escalation, resume, SDK JSON/NDJSON control events, status updates, rate-limit handling, model switch breadcrumbs, side questions.
 - Full settings parity gaps that remain: complete Zod-equivalent validation messages, full remote managed-settings watcher/non-daemon background refresh lifecycle beyond current turn-start and daemon-heartbeat refresh, remaining forked agent/frontmatter execution and UI edge cases beyond current policy metadata filtering, complete marketplace TUI/UI/background lifecycle beyond current local/settings/directory/file/URL catalog/git/github/npm cache manifest load, URL catalog cache fallback, git/github fetch/pull cache refresh, npm pack cache refresh, and background watcher/continuous app-state sync beyond current turn-start local settings reload.

internal/tool/executor.go

@@ -48,6 +48,7 @@ func (e Executor) Execute(ctx Context, use contracts.ToolUse, sink ProgressSink)
 		err := fmt.Errorf("%w: registry is nil", ErrUnknownTool)
 		return ErrorResult(use, err), err
 	}
+	ctx = e.withRegistryMetadata(ctx)
 	t, ok := e.Registry.Lookup(use.Name)
 	if !ok {
 		err := fmt.Errorf("%w: %s", ErrUnknownTool, use.Name)
@@ -143,6 +144,16 @@ func (e Executor) Execute(ctx Context, use contracts.ToolUse, sink ProgressSink)
 	return result, nil
 }
 
+func (e Executor) withRegistryMetadata(ctx Context) Context {
+	if ctx.Metadata == nil {
+		ctx.Metadata = map[string]any{}
+	}
+	if _, ok := ctx.Metadata[MetadataToolRegistryKey]; !ok && e.Registry != nil {
+		ctx.Metadata[MetadataToolRegistryKey] = e.Registry
+	}
+	return ctx
+}
+
 type defaultToolUseProgressSink struct {
 	sink      ProgressSink
 	toolUseID contracts.ID

internal/tool/types.go

@@ -80,6 +80,7 @@ const (
 	MetadataPolicySettingsKey      = "ccgo.policy_settings"
 	MetadataSessionPathKey         = "ccgo.session.path"
 	MetadataAvailableAgentsKey     = "ccgo.available_agents"
+	MetadataToolRegistryKey        = "ccgo.tool.registry"
 
 	HookPreToolUse        = "PreToolUse"
 	HookPostToolUse       = "PostToolUse"

internal/tools/file/builtin_tools_test.go

@@ -0,0 +1,12 @@
+package filetools
+
+import "testing"
+
+func TestBuiltinToolsIncludesToolSearch(t *testing.T) {
+	for _, tool := range BuiltinTools() {
+		if tool.Name() == "ToolSearch" {
+			return
+		}
+	}
+	t.Fatal("builtin tools did not include ToolSearch")
+}

internal/tools/file/tools.go

@@ -19,6 +19,7 @@ import (
 	"ccgo/internal/tool"
 	bashtools "ccgo/internal/tools/bash"
 	powershelltools "ccgo/internal/tools/powershell"
+	searchtools "ccgo/internal/tools/searchtools"
 	skilltools "ccgo/internal/tools/skill"
 	tasktools "ccgo/internal/tools/task"
 	todotools "ccgo/internal/tools/todo"
@@ -213,6 +214,7 @@ func BuiltinTools() []tool.Tool {
 		tasktools.NewScheduleCronTool(),
 		tasktools.NewRemoteTriggerTool(),
 		skilltools.NewSkillTool(),
+		searchtools.NewToolSearchTool(),
 		webtools.NewWebFetchTool(),
 		webtools.NewWebSearchTool(),
 	}