Update skill tool descriptions

docs/users/features/commands.md

@@ -59,6 +59,7 @@ Commands for managing AI tools and models.
 | ---------------- | --------------------------------------------- | --------------------------------------------- |
 | `/mcp`           | List configured MCP servers and tools         | `/mcp`, `/mcp desc`                           |
 | `/tools`         | Display currently available tool list         | `/tools`, `/tools desc`                       |
+| `/skills`        | List and run available skills (experimental)  | `/skills`, `/skills <name>`                   |
 | `/approval-mode` | Change approval mode for tool usage           | `/approval-mode <mode (auto-edit)> --project` |
 | →`plan`          | Analysis only, no execution                   | Secure review                                 |
 | →`default`       | Require approval for edits                    | Daily use                                     |

docs/users/features/skills.md

@@ -27,6 +27,14 @@ Agent Skills package expertise into discoverable capabilities. Each Skill consis
 
 Skills are **model-invoked** — the model autonomously decides when to use them based on your request and the Skill’s description. This is different from slash commands, which are **user-invoked** (you explicitly type `/command`).
 
+If you want to invoke a Skill explicitly, use the `/skills` slash command:
+
+```bash
+/skills <skill-name>
+```
+
+The `/skills` command is only available when you run with `--experimental-skills`. Use autocomplete to browse available Skills and descriptions.
+
 ### Benefits
 
 - Extend Qwen Code for your workflows

packages/cli/src/gemini.tsx

@@ -344,6 +344,7 @@ export async function main() {
       extensionEnablementManager,
       argv,
     );
+    registerCleanup(() => config.shutdown());
 
     if (config.getListExtensions()) {
       console.log('Installed extensions:');

packages/cli/src/services/BuiltinCommandLoader.ts

@@ -31,6 +31,7 @@ import { quitCommand } from '../ui/commands/quitCommand.js';
 import { restoreCommand } from '../ui/commands/restoreCommand.js';
 import { resumeCommand } from '../ui/commands/resumeCommand.js';
 import { settingsCommand } from '../ui/commands/settingsCommand.js';
+import { skillsCommand } from '../ui/commands/skillsCommand.js';
 import { statsCommand } from '../ui/commands/statsCommand.js';
 import { summaryCommand } from '../ui/commands/summaryCommand.js';
 import { terminalSetupCommand } from '../ui/commands/terminalSetupCommand.js';
@@ -78,6 +79,7 @@ export class BuiltinCommandLoader implements ICommandLoader {
       quitCommand,
       restoreCommand(this.config),
       resumeCommand,
+      ...(this.config?.getExperimentalSkills?.() ? [skillsCommand] : []),
       statsCommand,
       summaryCommand,
       themeCommand,

packages/cli/src/ui/commands/skillsCommand.ts

@@ -0,0 +1,132 @@
+/**
+ * @license
+ * Copyright 2025 Qwen
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import {
+  CommandKind,
+  type CommandCompletionItem,
+  type CommandContext,
+  type SlashCommand,
+} from './types.js';
+import { MessageType, type HistoryItemSkillsList } from '../types.js';
+import { t } from '../../i18n/index.js';
+import { AsyncFzf } from 'fzf';
+import type { SkillConfig } from '@qwen-code/qwen-code-core';
+
+export const skillsCommand: SlashCommand = {
+  name: 'skills',
+  get description() {
+    return t('List available skills.');
+  },
+  kind: CommandKind.BUILT_IN,
+  action: async (context: CommandContext, args?: string) => {
+    const rawArgs = args?.trim() ?? '';
+    const [skillName = ''] = rawArgs.split(/\s+/);
+
+    const skillManager = context.services.config?.getSkillManager();
+    if (!skillManager) {
+      context.ui.addItem(
+        {
+          type: MessageType.ERROR,
+          text: t('Could not retrieve skill manager.'),
+        },
+        Date.now(),
+      );
+      return;
+    }
+
+    const skills = await skillManager.listSkills();
+    if (skills.length === 0) {
+      context.ui.addItem(
+        {
+          type: MessageType.INFO,
+          text: t('No skills are currently available.'),
+        },
+        Date.now(),
+      );
+      return;
+    }
+
+    if (!skillName) {
+      const sortedSkills = [...skills].sort((left, right) =>
+        left.name.localeCompare(right.name),
+      );
+      const skillsListItem: HistoryItemSkillsList = {
+        type: MessageType.SKILLS_LIST,
+        skills: sortedSkills.map((skill) => ({ name: skill.name })),
+      };
+      context.ui.addItem(skillsListItem, Date.now());
+      return;
+    }
+    const normalizedName = skillName.toLowerCase();
+    const hasSkill = skills.some(
+      (skill) => skill.name.toLowerCase() === normalizedName,
+    );
+
+    if (!hasSkill) {
+      context.ui.addItem(
+        {
+          type: MessageType.ERROR,
+          text: t('Unknown skill: {{name}}', { name: skillName }),
+        },
+        Date.now(),
+      );
+      return;
+    }
+
+    const rawInput = context.invocation?.raw ?? `/skills ${rawArgs}`;
+    return {
+      type: 'submit_prompt',
+      content: [{ text: rawInput }],
+    };
+  },
+  completion: async (
+    context: CommandContext,
+    partialArg: string,
+  ): Promise<CommandCompletionItem[]> => {
+    const skillManager = context.services.config?.getSkillManager();
+    if (!skillManager) {
+      return [];
+    }
+
+    const skills = await skillManager.listSkills();
+    const normalizedPartial = partialArg.trim();
+    const matches = await getSkillMatches(skills, normalizedPartial);
+
+    return matches.map((skill) => ({
+      value: skill.name,
+      description: skill.description,
+    }));
+  },
+};
+
+async function getSkillMatches(
+  skills: SkillConfig[],
+  query: string,
+): Promise<SkillConfig[]> {
+  if (!query) {
+    return skills;
+  }
+
+  const names = skills.map((skill) => skill.name);
+  const skillMap = new Map(skills.map((skill) => [skill.name, skill]));
+
+  try {
+    const fzf = new AsyncFzf(names, {
+      fuzzy: 'v2',
+      casing: 'case-insensitive',
+    });
+    const results = (await fzf.find(query)) as Array<{ item: string }>;
+    return results
+      .map((result) => skillMap.get(result.item))
+      .filter((skill): skill is SkillConfig => !!skill);
+  } catch (error) {
+    console.error('[skillsCommand] Fuzzy match failed:', error);
+    const lowerQuery = query.toLowerCase();
+    return skills.filter((skill) =>
+      skill.name.toLowerCase().startsWith(lowerQuery),
+    );
+  }
+}

packages/cli/src/ui/commands/types.ts

@@ -209,6 +209,12 @@ export enum CommandKind {
   MCP_PROMPT = 'mcp-prompt',
 }
 
+export interface CommandCompletionItem {
+  value: string;
+  label?: string;
+  description?: string;
+}
+
 // The standardized contract for any command in the system.
 export interface SlashCommand {
   name: string;
@@ -234,7 +240,7 @@ export interface SlashCommand {
   completion?: (
     context: CommandContext,
     partialArg: string,
-  ) => Promise<string[]>;
+  ) => Promise<Array<string | CommandCompletionItem> | null>;
 
   subCommands?: SlashCommand[];
 }

packages/cli/src/ui/components/HistoryItemDisplay.tsx

@@ -30,6 +30,7 @@ import { Help } from './Help.js';
 import type { SlashCommand } from '../commands/types.js';
 import { ExtensionsList } from './views/ExtensionsList.js';
 import { getMCPServerStatus } from '@qwen-code/qwen-code-core';
+import { SkillsList } from './views/SkillsList.js';
 import { ToolsList } from './views/ToolsList.js';
 import { McpStatus } from './views/McpStatus.js';
 
@@ -153,6 +154,9 @@ const HistoryItemDisplayComponent: React.FC<HistoryItemDisplayProps> = ({
           showDescriptions={itemForDisplay.showDescriptions}
         />
       )}
+      {itemForDisplay.type === 'skills_list' && (
+        <SkillsList skills={itemForDisplay.skills} />
+      )}
       {itemForDisplay.type === 'mcp_status' && (
         <McpStatus {...itemForDisplay} serverStatus={getMCPServerStatus} />
       )}

packages/cli/src/ui/components/SuggestionsDisplay.tsx

@@ -106,7 +106,7 @@ export function SuggestionsDisplay({
             </Box>
 
             {suggestion.description && (
-              <Box flexGrow={1} paddingLeft={3}>
+              <Box flexGrow={1} paddingLeft={2}>
                 <Text color={textColor} wrap="truncate">
                   {suggestion.description}
                 </Text>

packages/cli/src/ui/components/messages/InfoMessage.tsx

@@ -23,7 +23,7 @@ export const InfoMessage: React.FC<InfoMessageProps> = ({ text }) => {
   const prefixWidth = prefix.length;
 
   return (
-    <Box flexDirection="row" marginTop={1}>
+    <Box flexDirection="row" marginBottom={1}>
       <Box width={prefixWidth}>
         <Text color={theme.status.warning}>{prefix}</Text>
       </Box>

packages/cli/src/ui/components/messages/WarningMessage.tsx

@@ -18,7 +18,7 @@ export const WarningMessage: React.FC<WarningMessageProps> = ({ text }) => {
   const prefixWidth = 3;
 
   return (
-    <Box flexDirection="row" marginTop={1}>
+    <Box flexDirection="row" marginBottom={1}>
       <Box width={prefixWidth}>
         <Text color={Colors.AccentYellow}>{prefix}</Text>
       </Box>

packages/cli/src/ui/components/views/SkillsList.tsx

@@ -0,0 +1,36 @@
+/**
+ * @license
+ * Copyright 2025 Qwen
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type React from 'react';
+import { Box, Text } from 'ink';
+import { theme } from '../../semantic-colors.js';
+import { type SkillDefinition } from '../../types.js';
+import { t } from '../../../i18n/index.js';
+
+interface SkillsListProps {
+  skills: readonly SkillDefinition[];
+}
+
+export const SkillsList: React.FC<SkillsListProps> = ({ skills }) => (
+  <Box flexDirection="column" marginBottom={1}>
+    <Text bold color={theme.text.primary}>
+      {t('Available skills:')}
+    </Text>
+    <Box height={1} />
+    {skills.length > 0 ? (
+      skills.map((skill) => (
+        <Box key={skill.name} flexDirection="row">
+          <Text color={theme.text.primary}>{'  '}- </Text>
+          <Text bold color={theme.text.accent}>
+            {skill.name}
+          </Text>
+        </Box>
+      ))
+    ) : (
+      <Text color={theme.text.primary}> {t('No skills available')}</Text>
+    )}
+  </Box>
+);

packages/cli/src/ui/hooks/useSlashCompletion.test.ts

@@ -573,6 +573,45 @@ describe('useSlashCompletion', () => {
       });
     });
 
+    it('should map completion items with descriptions for argument suggestions', async () => {
+      const mockCompletionFn = vi.fn().mockResolvedValue([
+        { value: 'pdf', description: 'Create PDF documents' },
+        { value: 'xlsx', description: 'Work with spreadsheets' },
+      ]);
+
+      const slashCommands = [
+        createTestCommand({
+          name: 'skills',
+          description: 'List available skills',
+          completion: mockCompletionFn,
+        }),
+      ];
+
+      const { result } = renderHook(() =>
+        useTestHarnessForSlashCompletion(
+          true,
+          '/skills ',
+          slashCommands,
+          mockCommandContext,
+        ),
+      );
+
+      await waitFor(() => {
+        expect(result.current.suggestions).toEqual([
+          {
+            label: 'pdf',
+            value: 'pdf',
+            description: 'Create PDF documents',
+          },
+          {
+            label: 'xlsx',
+            value: 'xlsx',
+            description: 'Work with spreadsheets',
+          },
+        ]);
+      });
+    });
+
     it('should call command.completion with an empty string when args start with a space', async () => {
       const mockCompletionFn = vi
         .fn()

packages/cli/src/ui/hooks/useSlashCompletion.ts

@@ -9,6 +9,7 @@ import { AsyncFzf } from 'fzf';
 import type { Suggestion } from '../components/SuggestionsDisplay.js';
 import {
   CommandKind,
+  type CommandCompletionItem,
   type CommandContext,
   type SlashCommand,
 } from '../commands/types.js';
@@ -215,10 +216,9 @@ function useCommandSuggestions(
             )) || [];
 
           if (!signal.aborted) {
-            const finalSuggestions = results.map((s) => ({
-              label: s,
-              value: s,
-            }));
+            const finalSuggestions = results
+              .map((item) => toSuggestion(item))
+              .filter((suggestion): suggestion is Suggestion => !!suggestion);
             setSuggestions(finalSuggestions);
             setIsLoading(false);
           }
@@ -310,6 +310,20 @@ function useCommandSuggestions(
   return { suggestions, isLoading };
 }
 
+function toSuggestion(item: string | CommandCompletionItem): Suggestion | null {
+  if (typeof item === 'string') {
+    return { label: item, value: item };
+  }
+  if (!item.value) {
+    return null;
+  }
+  return {
+    label: item.label ?? item.value,
+    value: item.value,
+    description: item.description,
+  };
+}
+
 function useCompletionPositions(
   query: string | null,
   parserResult: CommandParserResult,

packages/cli/src/ui/types.ts

@@ -201,12 +201,21 @@ export interface ToolDefinition {
   description?: string;
 }
 
+export interface SkillDefinition {
+  name: string;
+}
+
 export type HistoryItemToolsList = HistoryItemBase & {
   type: 'tools_list';
   tools: ToolDefinition[];
   showDescriptions: boolean;
 };
 
+export type HistoryItemSkillsList = HistoryItemBase & {
+  type: 'skills_list';
+  skills: SkillDefinition[];
+};
+
 // JSON-friendly types for using as a simple data model showing info about an
 // MCP Server.
 export interface JsonMcpTool {
@@ -268,6 +277,7 @@ export type HistoryItemWithoutId =
   | HistoryItemCompression
   | HistoryItemExtensionsList
   | HistoryItemToolsList
+  | HistoryItemSkillsList
   | HistoryItemMcpStatus;
 
 export type HistoryItem = HistoryItemWithoutId & { id: number };
@@ -289,6 +299,7 @@ export enum MessageType {
   SUMMARY = 'summary',
   EXTENSIONS_LIST = 'extensions_list',
   TOOLS_LIST = 'tools_list',
+  SKILLS_LIST = 'skills_list',
   MCP_STATUS = 'mcp_status',
 }
 

packages/core/src/config/config.ts

@@ -650,6 +650,7 @@ export class Config {
     this.promptRegistry = new PromptRegistry();
     this.subagentManager = new SubagentManager(this);
     this.skillManager = new SkillManager(this);
+    await this.skillManager.startWatching();
 
     // Load session subagents if they were provided before initialization
     if (this.sessionSubagents.length > 0) {
@@ -734,6 +735,13 @@ export class Config {
     return this.sessionId;
   }
 
+  /**
+   * Releases resources owned by the config instance.
+   */
+  async shutdown(): Promise<void> {
+    this.skillManager?.stopWatching();
+  }
+
   /**
    * Starts a new session and resets session-scoped services.
    */

packages/core/src/skills/skill-manager.ts

@@ -5,6 +5,7 @@
  */
 
 import * as fs from 'fs/promises';
+import * as fsSync from 'fs';
 import * as path from 'path';
 import * as os from 'os';
 import { parse as parseYaml } from '../utils/yaml-parser.js';
@@ -29,6 +30,9 @@ export class SkillManager {
   private skillsCache: Map<SkillLevel, SkillConfig[]> | null = null;
   private readonly changeListeners: Set<() => void> = new Set();
   private parseErrors: Map<string, SkillError> = new Map();
+  private readonly watchers: Map<string, fsSync.FSWatcher> = new Map();
+  private watchStarted = false;
+  private refreshTimer: NodeJS.Timeout | null = null;
 
   constructor(private readonly config: Config) {}
 
@@ -221,6 +225,34 @@ export class SkillManager {
     this.notifyChangeListeners();
   }
 
+  /**
+   * Starts watching skill directories for changes.
+   */
+  async startWatching(): Promise<void> {
+    if (this.watchStarted) {
+      return;
+    }
+
+    this.watchStarted = true;
+    await this.refreshCache();
+    this.updateWatchersFromCache();
+  }
+
+  /**
+   * Stops watching skill directories for changes.
+   */
+  stopWatching(): void {
+    for (const watcher of this.watchers.values()) {
+      watcher.close();
+    }
+    this.watchers.clear();
+    this.watchStarted = false;
+    if (this.refreshTimer) {
+      clearTimeout(this.refreshTimer);
+      this.refreshTimer = null;
+    }
+  }
+
   /**
    * Parses a SKILL.md file and returns the configuration.
    *
@@ -449,4 +481,69 @@ export class SkillManager {
       this.skillsCache.set(level, levelSkills);
     }
   }
+
+  private updateWatchersFromCache(): void {
+    const desiredPaths = new Set<string>();
+    const recursiveSupported =
+      process.platform === 'darwin' || process.platform === 'win32';
+
+    for (const level of ['project', 'user'] as const) {
+      const baseDir = this.getSkillsBaseDir(level);
+      const parentDir = path.dirname(baseDir);
+      if (fsSync.existsSync(parentDir)) {
+        desiredPaths.add(parentDir);
+      }
+      if (fsSync.existsSync(baseDir)) {
+        desiredPaths.add(baseDir);
+      }
+
+      const levelSkills = this.skillsCache?.get(level) || [];
+      for (const skill of levelSkills) {
+        const skillDir = path.dirname(skill.filePath);
+        if (fsSync.existsSync(skillDir)) {
+          desiredPaths.add(skillDir);
+        }
+      }
+    }
+
+    for (const existingPath of this.watchers.keys()) {
+      if (!desiredPaths.has(existingPath)) {
+        this.watchers.get(existingPath)?.close();
+        this.watchers.delete(existingPath);
+      }
+    }
+
+    for (const watchPath of desiredPaths) {
+      if (this.watchers.has(watchPath)) {
+        continue;
+      }
+
+      try {
+        const watcher = fsSync.watch(
+          watchPath,
+          { recursive: recursiveSupported },
+          () => {
+            this.scheduleRefresh();
+          },
+        );
+        this.watchers.set(watchPath, watcher);
+      } catch (error) {
+        console.warn(
+          `Failed to watch skills directory at ${watchPath}:`,
+          error,
+        );
+      }
+    }
+  }
+
+  private scheduleRefresh(): void {
+    if (this.refreshTimer) {
+      clearTimeout(this.refreshTimer);
+    }
+
+    this.refreshTimer = setTimeout(() => {
+      this.refreshTimer = null;
+      void this.refreshCache().then(() => this.updateWatchersFromCache());
+    }, 150);
+  }
 }

packages/core/src/tools/__snapshots__/shell.test.ts.snap

@@ -2,66 +2,88 @@
 
 exports[`ShellTool > getDescription > should return the non-windows description when not on windows 1`] = `
 "This tool executes a given shell command as \`bash -c <command>\`. Command can start background processes using \`&\`. Command is executed as a subprocess that leads its own process group. Command process group can be terminated as \`kill -- -PGID\` or signaled as \`kill -s SIGNAL -- -PGID\`.
+**Usage notes**:
+- The command argument is required.
+- It is very helpful if you write a clear, concise description of what this command does in 5-10 words.
 
-      **Background vs Foreground Execution:**
-      You should decide whether commands should run in background or foreground based on their nature:
-      
-      **Use background execution (is_background: true) for:**
-      - Long-running development servers: \`npm run start\`, \`npm run dev\`, \`yarn dev\`, \`bun run start\`
-      - Build watchers: \`npm run watch\`, \`webpack --watch\`
-      - Database servers: \`mongod\`, \`mysql\`, \`redis-server\`
-      - Web servers: \`python -m http.server\`, \`php -S localhost:8000\`
-      - Any command expected to run indefinitely until manually stopped
-      
-      **Use foreground execution (is_background: false) for:**
-      - One-time commands: \`ls\`, \`cat\`, \`grep\`
-      - Build commands: \`npm run build\`, \`make\`
-      - Installation commands: \`npm install\`, \`pip install\`
-      - Git operations: \`git commit\`, \`git push\`
-      - Test runs: \`npm test\`, \`pytest\`
-      
-      The following information is returned:
+- Avoid using run_shell_command with the \`find\`, \`grep\`, \`cat\`, \`head\`, \`tail\`, \`sed\`, \`awk\`, or \`echo\` commands, unless explicitly instructed or when these commands are truly necessary for the task. Instead, always prefer using the dedicated tools for these commands:
+  - File search: Use glob (NOT find or ls)
+  - Content search: Use grep_search (NOT grep or rg)
+  - Read files: Use read_file (NOT cat/head/tail)
+  - Edit files: Use edit (NOT sed/awk)
+  - Write files: Use write_file (NOT echo >/cat <<EOF)
+  - Communication: Output text directly (NOT echo/printf)
+- When issuing multiple commands:
+  - If the commands are independent and can run in parallel, make multiple run_shell_command tool calls in a single message. For example, if you need to run "git status" and "git diff", send a single message with two run_shell_command tool calls in parallel.
+  - If the commands depend on each other and must run sequentially, use a single run_shell_command call with '&&' to chain them together (e.g., \`git add . && git commit -m "message" && git push\`). For instance, if one operation must complete before another starts (like mkdir before cp, Write before run_shell_command for git operations, or git add before git commit), run these operations sequentially instead.
+  - Use ';' only when you need to run commands sequentially but don't care if earlier commands fail
+  - DO NOT use newlines to separate commands (newlines are ok in quoted strings)
+- Try to maintain your current working directory throughout the session by using absolute paths and avoiding usage of \`cd\`. You may use \`cd\` if the User explicitly requests it.
+  <good-example>
+  pytest /foo/bar/tests
+  </good-example>
+  <bad-example>
+  cd /foo/bar && pytest tests
+  </bad-example>
 
-      Command: Executed command.
-      Directory: Directory where command was executed, or \`(root)\`.
-      Stdout: Output on stdout stream. Can be \`(empty)\` or partial on error and for any unwaited background processes.
-      Stderr: Output on stderr stream. Can be \`(empty)\` or partial on error and for any unwaited background processes.
-      Error: Error or \`(none)\` if no error was reported for the subprocess.
-      Exit Code: Exit code or \`(none)\` if terminated by signal.
-      Signal: Signal number or \`(none)\` if no signal was received.
-      Background PIDs: List of background processes started or \`(none)\`.
-      Process Group PGID: Process group started or \`(none)\`"
+**Background vs Foreground Execution:**
+You should decide whether commands should run in background or foreground based on their nature:
+
+**Use background execution (is_background: true) for:**
+- Long-running development servers: \`npm run start\`, \`npm run dev\`, \`yarn dev\`, \`bun run start\`
+- Build watchers: \`npm run watch\`, \`webpack --watch\`
+- Database servers: \`mongod\`, \`mysql\`, \`redis-server\`
+- Web servers: \`python -m http.server\`, \`php -S localhost:8000\`
+- Any command expected to run indefinitely until manually stopped
+
+**Use foreground execution (is_background: false) for:**
+- One-time commands: \`ls\`, \`cat\`, \`grep\`
+- Build commands: \`npm run build\`, \`make\`
+- Installation commands: \`npm install\`, \`pip install\`
+- Git operations: \`git commit\`, \`git push\`
+- Test runs: \`npm test\`, \`pytest\`"
 `;
 
 exports[`ShellTool > getDescription > should return the windows description when on windows 1`] = `
 "This tool executes a given shell command as \`cmd.exe /c <command>\`. Command can start background processes using \`start /b\`.
+**Usage notes**:
+- The command argument is required.
+- It is very helpful if you write a clear, concise description of what this command does in 5-10 words.
 
-      **Background vs Foreground Execution:**
-      You should decide whether commands should run in background or foreground based on their nature:
-      
-      **Use background execution (is_background: true) for:**
-      - Long-running development servers: \`npm run start\`, \`npm run dev\`, \`yarn dev\`, \`bun run start\`
-      - Build watchers: \`npm run watch\`, \`webpack --watch\`
-      - Database servers: \`mongod\`, \`mysql\`, \`redis-server\`
-      - Web servers: \`python -m http.server\`, \`php -S localhost:8000\`
-      - Any command expected to run indefinitely until manually stopped
-      
-      **Use foreground execution (is_background: false) for:**
-      - One-time commands: \`ls\`, \`cat\`, \`grep\`
-      - Build commands: \`npm run build\`, \`make\`
-      - Installation commands: \`npm install\`, \`pip install\`
-      - Git operations: \`git commit\`, \`git push\`
-      - Test runs: \`npm test\`, \`pytest\`
-      
-      The following information is returned:
+- Avoid using run_shell_command with the \`find\`, \`grep\`, \`cat\`, \`head\`, \`tail\`, \`sed\`, \`awk\`, or \`echo\` commands, unless explicitly instructed or when these commands are truly necessary for the task. Instead, always prefer using the dedicated tools for these commands:
+  - File search: Use glob (NOT find or ls)
+  - Content search: Use grep_search (NOT grep or rg)
+  - Read files: Use read_file (NOT cat/head/tail)
+  - Edit files: Use edit (NOT sed/awk)
+  - Write files: Use write_file (NOT echo >/cat <<EOF)
+  - Communication: Output text directly (NOT echo/printf)
+- When issuing multiple commands:
+  - If the commands are independent and can run in parallel, make multiple run_shell_command tool calls in a single message. For example, if you need to run "git status" and "git diff", send a single message with two run_shell_command tool calls in parallel.
+  - If the commands depend on each other and must run sequentially, use a single run_shell_command call with '&&' to chain them together (e.g., \`git add . && git commit -m "message" && git push\`). For instance, if one operation must complete before another starts (like mkdir before cp, Write before run_shell_command for git operations, or git add before git commit), run these operations sequentially instead.
+  - Use ';' only when you need to run commands sequentially but don't care if earlier commands fail
+  - DO NOT use newlines to separate commands (newlines are ok in quoted strings)
+- Try to maintain your current working directory throughout the session by using absolute paths and avoiding usage of \`cd\`. You may use \`cd\` if the User explicitly requests it.
+  <good-example>
+  pytest /foo/bar/tests
+  </good-example>
+  <bad-example>
+  cd /foo/bar && pytest tests
+  </bad-example>
 
-      Command: Executed command.
-      Directory: Directory where command was executed, or \`(root)\`.
-      Stdout: Output on stdout stream. Can be \`(empty)\` or partial on error and for any unwaited background processes.
-      Stderr: Output on stderr stream. Can be \`(empty)\` or partial on error and for any unwaited background processes.
-      Error: Error or \`(none)\` if no error was reported for the subprocess.
-      Exit Code: Exit code or \`(none)\` if terminated by signal.
-      Signal: Signal number or \`(none)\` if no signal was received.
-      Background PIDs: List of background processes started or \`(none)\`.
-      Process Group PGID: Process group started or \`(none)\`"
+**Background vs Foreground Execution:**
+You should decide whether commands should run in background or foreground based on their nature:
+
+**Use background execution (is_background: true) for:**
+- Long-running development servers: \`npm run start\`, \`npm run dev\`, \`yarn dev\`, \`bun run start\`
+- Build watchers: \`npm run watch\`, \`webpack --watch\`
+- Database servers: \`mongod\`, \`mysql\`, \`redis-server\`
+- Web servers: \`python -m http.server\`, \`php -S localhost:8000\`
+- Any command expected to run indefinitely until manually stopped
+
+**Use foreground execution (is_background: false) for:**
+- One-time commands: \`ls\`, \`cat\`, \`grep\`
+- Build commands: \`npm run build\`, \`make\`
+- Installation commands: \`npm install\`, \`pip install\`
+- Git operations: \`git commit\`, \`git push\`
+- Test runs: \`npm test\`, \`pytest\`"
 `;

packages/core/src/tools/shell.test.ts

@@ -59,6 +59,9 @@ describe('ShellTool', () => {
       getWorkspaceContext: vi
         .fn()
         .mockReturnValue(createMockWorkspaceContext('/test/dir')),
+      storage: {
+        getUserSkillsDir: vi.fn().mockReturnValue('/test/dir/.qwen/skills'),
+      },
       getGeminiClient: vi.fn(),
       getGitCoAuthor: vi.fn().mockReturnValue({
         enabled: true,
@@ -142,6 +145,42 @@ describe('ShellTool', () => {
       );
     });
 
+    it('should throw an error for a directory within the user skills directory', () => {
+      expect(() =>
+        shellTool.build({
+          command: 'ls',
+          directory: '/test/dir/.qwen/skills/my-skill',
+          is_background: false,
+        }),
+      ).toThrow(
+        'Explicitly running shell commands from within the user skills directory is not allowed. Please use absolute paths for command parameter instead.',
+      );
+    });
+
+    it('should throw an error for the user skills directory itself', () => {
+      expect(() =>
+        shellTool.build({
+          command: 'ls',
+          directory: '/test/dir/.qwen/skills',
+          is_background: false,
+        }),
+      ).toThrow(
+        'Explicitly running shell commands from within the user skills directory is not allowed. Please use absolute paths for command parameter instead.',
+      );
+    });
+
+    it('should resolve directory path before checking user skills directory', () => {
+      expect(() =>
+        shellTool.build({
+          command: 'ls',
+          directory: '/test/dir/.qwen/skills/../skills/my-skill',
+          is_background: false,
+        }),
+      ).toThrow(
+        'Explicitly running shell commands from within the user skills directory is not allowed. Please use absolute paths for command parameter instead.',
+      );
+    });
+
     it('should return an invocation for a valid absolute directory path', () => {
       (mockConfig.getWorkspaceContext as Mock).mockReturnValue(
         createMockWorkspaceContext('/test/dir', ['/another/workspace']),

packages/core/src/tools/shell.ts

@@ -34,6 +34,7 @@ import type {
 import { ShellExecutionService } from '../services/shellExecutionService.js';
 import { formatMemoryUsage } from '../utils/formatters.js';
 import type { AnsiOutput } from '../utils/terminalSerializer.js';
+import { isSubpath } from '../utils/paths.js';
 import {
   getCommandRoots,
   isCommandAllowed,
@@ -407,35 +408,46 @@ Co-authored-by: ${gitCoAuthorSettings.name} <${gitCoAuthorSettings.email}>`;
 
 function getShellToolDescription(): string {
   const toolDescription = `
+**Usage notes**:
+- The command argument is required.
+- It is very helpful if you write a clear, concise description of what this command does in 5-10 words.
 
-      **Background vs Foreground Execution:**
-      You should decide whether commands should run in background or foreground based on their nature:
-      
-      **Use background execution (is_background: true) for:**
-      - Long-running development servers: \`npm run start\`, \`npm run dev\`, \`yarn dev\`, \`bun run start\`
-      - Build watchers: \`npm run watch\`, \`webpack --watch\`
-      - Database servers: \`mongod\`, \`mysql\`, \`redis-server\`
-      - Web servers: \`python -m http.server\`, \`php -S localhost:8000\`
-      - Any command expected to run indefinitely until manually stopped
-      
-      **Use foreground execution (is_background: false) for:**
-      - One-time commands: \`ls\`, \`cat\`, \`grep\`
-      - Build commands: \`npm run build\`, \`make\`
-      - Installation commands: \`npm install\`, \`pip install\`
-      - Git operations: \`git commit\`, \`git push\`
-      - Test runs: \`npm test\`, \`pytest\`
-      
-      The following information is returned:
+- Avoid using run_shell_command with the \`find\`, \`grep\`, \`cat\`, \`head\`, \`tail\`, \`sed\`, \`awk\`, or \`echo\` commands, unless explicitly instructed or when these commands are truly necessary for the task. Instead, always prefer using the dedicated tools for these commands:
+  - File search: Use glob (NOT find or ls)
+  - Content search: Use grep_search (NOT grep or rg)
+  - Read files: Use read_file (NOT cat/head/tail)
+  - Edit files: Use edit (NOT sed/awk)
+  - Write files: Use write_file (NOT echo >/cat <<EOF)
+  - Communication: Output text directly (NOT echo/printf)
+- When issuing multiple commands:
+  - If the commands are independent and can run in parallel, make multiple run_shell_command tool calls in a single message. For example, if you need to run "git status" and "git diff", send a single message with two run_shell_command tool calls in parallel.
+  - If the commands depend on each other and must run sequentially, use a single run_shell_command call with '&&' to chain them together (e.g., \`git add . && git commit -m "message" && git push\`). For instance, if one operation must complete before another starts (like mkdir before cp, Write before run_shell_command for git operations, or git add before git commit), run these operations sequentially instead.
+  - Use ';' only when you need to run commands sequentially but don't care if earlier commands fail
+  - DO NOT use newlines to separate commands (newlines are ok in quoted strings)
+- Try to maintain your current working directory throughout the session by using absolute paths and avoiding usage of \`cd\`. You may use \`cd\` if the User explicitly requests it.
+  <good-example>
+  pytest /foo/bar/tests
+  </good-example>
+  <bad-example>
+  cd /foo/bar && pytest tests
+  </bad-example>
 
-      Command: Executed command.
-      Directory: Directory where command was executed, or \`(root)\`.
-      Stdout: Output on stdout stream. Can be \`(empty)\` or partial on error and for any unwaited background processes.
-      Stderr: Output on stderr stream. Can be \`(empty)\` or partial on error and for any unwaited background processes.
-      Error: Error or \`(none)\` if no error was reported for the subprocess.
-      Exit Code: Exit code or \`(none)\` if terminated by signal.
-      Signal: Signal number or \`(none)\` if no signal was received.
-      Background PIDs: List of background processes started or \`(none)\`.
-      Process Group PGID: Process group started or \`(none)\``;
+**Background vs Foreground Execution:**
+You should decide whether commands should run in background or foreground based on their nature:
+
+**Use background execution (is_background: true) for:**
+- Long-running development servers: \`npm run start\`, \`npm run dev\`, \`yarn dev\`, \`bun run start\`
+- Build watchers: \`npm run watch\`, \`webpack --watch\`
+- Database servers: \`mongod\`, \`mysql\`, \`redis-server\`
+- Web servers: \`python -m http.server\`, \`php -S localhost:8000\`
+- Any command expected to run indefinitely until manually stopped
+
+**Use foreground execution (is_background: false) for:**
+- One-time commands: \`ls\`, \`cat\`, \`grep\`
+- Build commands: \`npm run build\`, \`make\`
+- Installation commands: \`npm install\`, \`pip install\`
+- Git operations: \`git commit\`, \`git push\`
+- Test runs: \`npm test\`, \`pytest\``;
 
   if (os.platform() === 'win32') {
     return `This tool executes a given shell command as \`cmd.exe /c <command>\`. Command can start background processes using \`start /b\`.${toolDescription}`;
@@ -526,6 +538,17 @@ export class ShellTool extends BaseDeclarativeTool<
       if (!path.isAbsolute(params.directory)) {
         return 'Directory must be an absolute path.';
       }
+
+      const userSkillsDir = this.config.storage.getUserSkillsDir();
+      const resolvedDirectoryPath = path.resolve(params.directory);
+      const isWithinUserSkills = isSubpath(
+        userSkillsDir,
+        resolvedDirectoryPath,
+      );
+      if (isWithinUserSkills) {
+        return `Explicitly running shell commands from within the user skills directory is not allowed. Please use absolute paths for command parameter instead.`;
+      }
+
       const workspaceDirs = this.config.getWorkspaceContext().getDirectories();
       const isWithinWorkspace = workspaceDirs.some((wsDir) =>
         params.directory!.startsWith(wsDir),

packages/core/src/tools/skill.test.ts

@@ -324,7 +324,9 @@ describe('SkillTool', () => {
         'Review code for quality and best practices.',
       );
 
-      expect(result.returnDisplay).toBe('Launching skill: code-review');
+      expect(result.returnDisplay).toBe(
+        'Specialized skill for reviewing code quality',
+      );
     });
 
     it('should include allowedTools in result when present', async () => {
@@ -349,7 +351,7 @@ describe('SkillTool', () => {
       // Base description is omitted from llmContent; ensure body is present.
       expect(llmText).toContain('Help write comprehensive tests.');
 
-      expect(result.returnDisplay).toBe('Launching skill: testing');
+      expect(result.returnDisplay).toBe('Skill for writing and running tests');
     });
 
     it('should handle skill not found error', async () => {
@@ -416,7 +418,7 @@ describe('SkillTool', () => {
       ).createInvocation(params);
       const description = invocation.getDescription();
 
-      expect(description).toBe('Launching skill: "code-review"');
+      expect(description).toBe('Use skill: "code-review"');
     });
 
     it('should handle skill without additional files', async () => {
@@ -436,7 +438,9 @@ describe('SkillTool', () => {
       const llmText = partToString(result.llmContent);
       expect(llmText).not.toContain('## Additional Files');
 
-      expect(result.returnDisplay).toBe('Launching skill: code-review');
+      expect(result.returnDisplay).toBe(
+        'Specialized skill for reviewing code quality',
+      );
     });
   });
 });

packages/core/src/tools/skill.ts

@@ -49,7 +49,7 @@ export class SkillTool extends BaseDeclarativeTool<SkillParams, ToolResult> {
       'Execute a skill within the main conversation. Loading available skills...', // Initial description
       Kind.Read,
       initialSchema,
-      true, // isOutputMarkdown
+      false, // isOutputMarkdown
       false, // canUpdateOutput
     );
 
@@ -128,6 +128,10 @@ Important:
 - Only use skills listed in <available_skills> below
 - Do not invoke a skill that is already running
 - Do not use this tool for built-in CLI commands (like /help, /clear, etc.)
+- When executing scripts or loading referenced files, ALWAYS resolve absolute paths from skill's base directory. Examples:
+  - \`bash scripts/init.sh\` -> \`bash /path/to/skill/scripts/init.sh\`
+  - \`python scripts/helper.py\` -> \`python /path/to/skill/scripts/helper.py\`
+  - \`reference.md\` -> \`/path/to/skill/reference.md\`
 </skills_instructions>
 
 <available_skills>
@@ -183,7 +187,7 @@ class SkillToolInvocation extends BaseToolInvocation<SkillParams, ToolResult> {
   }
 
   getDescription(): string {
-    return `Launching skill: "${this.params.skill}"`;
+    return `Use skill: "${this.params.skill}"`;
   }
 
   override async shouldConfirmExecute(): Promise<false> {
@@ -238,16 +242,16 @@ class SkillToolInvocation extends BaseToolInvocation<SkillParams, ToolResult> {
       const baseDir = path.dirname(skill.filePath);
 
       // Build markdown content for LLM (show base dir, then body)
-      const llmContent = `Base directory for this skill: ${baseDir}\n\n${skill.body}\n`;
+      const llmContent = `Base directory for this skill: ${baseDir}\nImportant: ALWAYS resolve absolute paths from this base directory when working with skills.\n\n${skill.body}\n`;
 
       return {
         llmContent: [{ text: llmContent }],
-        returnDisplay: `Launching skill: ${skill.name}`,
+        returnDisplay: skill.description,
       };
     } catch (error) {
       const errorMessage =
         error instanceof Error ? error.message : String(error);
-      console.error(`[SkillsTool] Error launching skill: ${errorMessage}`);
+      console.error(`[SkillsTool] Error using skill: ${errorMessage}`);
 
       // Log failed skill launch
       logSkillLaunch(