@shawnstack/quickforge 1.3.18 → 1.3.20

package/server/tools/index.mjs

@@ -1,911 +1,1013 @@
-import { createWriteStream, promises as fs } from 'node:fs'
-import path from 'node:path'
-import { spawn } from 'node:child_process'
-import { createRequire } from 'node:module'
-import { resolveWorkspacePath, toWorkspaceRelative, assertSafeWorkspacePath, truncateText, splitLines, walkFiles } from '../utils/workspace.mjs'
-import { logsDir } from '../storage.mjs'
-import { createTextDiff } from '../utils/text-diff.mjs'
-import {
-  formatSkillActivation,
-  loadSelectedGlobalSkills,
-  loadSelectedProjectSkills,
-  mergeSkills,
-  readSkillResource,
-} from '../skills.mjs'
-import { getToolWorkspaceRoot } from '../utils/workspace.mjs'
-
-const require = createRequire(import.meta.url)
-
-// --- read_file ---
-export async function toolReadFile(params, context) {
-  const file = resolveWorkspacePath(params?.path, context)
-  await assertSafeWorkspacePath(file, context)
-
-  const text = await fs.readFile(file, 'utf8')
-  const lines = splitLines(text)
-  const offset = Math.max(1, Number(params?.offset || 1))
-  const limit = Math.min(2000, Math.max(1, Number(params?.limit || 200)))
-  const selected = lines.slice(offset - 1, offset - 1 + limit)
-  const content = selected.map((line, index) => `${offset + index}: ${line}`).join('\n')
-  const suffix = offset - 1 + limit < lines.length ? `\n\n[showing ${selected.length} of ${lines.length} lines]` : ''
-
-  return {
-    content: truncateText(`${content}${suffix}`),
-    details: { path: toWorkspaceRelative(file, context), project: context?.project, totalLines: lines.length, offset, limit },
-  }
-}
-
-// --- grep_files ---
-
-const RIPGREP_MAX_FILESIZE = '1M'
-const RIPGREP_TIMEOUT_MS = 60 * 1000
-const DEFAULT_EXCLUDE_GLOBS = [
-  '!.git/**',
-  '!node_modules/**',
-  '!dist/**',
-  '!dist-ssr/**',
-  '!.vite/**',
-  '!**/*.png',
-  '!**/*.jpg',
-  '!**/*.jpeg',
-  '!**/*.gif',
-  '!**/*.webp',
-  '!**/*.ico',
-  '!**/*.pdf',
-  '!**/*.zip',
-  '!**/*.gz',
-  '!**/*.7z',
-  '!**/*.exe',
-  '!**/*.dll',
-  '!**/*.woff',
-  '!**/*.woff2',
-  '!**/*.ttf',
-]
-const SENSITIVE_EXCLUDE_GLOBS = [
-  '!.env',
-  '!**/.env',
-  '!.env.*',
-  '!**/.env.*',
-  '!**/*.pem',
-  '!**/*.key',
-  '!**/*.p12',
-  '!**/*.pfx',
-  '!**/*.crt',
-  '!**/*.cer',
-  '!**/*.token',
-  '!credentials.json',
-  '!**/credentials.json',
-  '!secrets.json',
-  '!**/secrets.json',
-  '!id_rsa',
-  '!**/id_rsa',
-  '!id_ed25519',
-  '!**/id_ed25519',
-]
-
-let cachedRipgrepExecutable
-
-/**
- * Process items with bounded concurrency.  Returns results in input order.
- * @template T, R
- * @param {T[]} items
- * @param {(item: T, index: number) => Promise<R>} fn
- * @param {number} concurrency
- * @returns {Promise<R[]>}
- */
-async function poolMap(items, fn, concurrency = 20) {
-  const results = new Array(items.length)
-  let cursor = 0
-
-  async function worker() {
-    while (cursor < items.length) {
-      const index = cursor++
-      results[index] = await fn(items[index], index)
-    }
-  }
-
-  const workers = Array.from({ length: Math.min(concurrency, items.length) }, () => worker())
-  await Promise.all(workers)
-  return results
-}
-
-function clampNumber(value, defaultValue, min, max) {
-  const number = Number(value)
-  if (!Number.isFinite(number)) return defaultValue
-  return Math.min(max, Math.max(min, Math.trunc(number)))
-}
-
-function escapeRegExp(value) {
-  return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')
-}
-
-function normalizeGlobList(value) {
-  const values = Array.isArray(value) ? value : typeof value === 'string' ? [value] : []
-  return values
-    .map((item) => String(item || '').trim())
-    .filter(Boolean)
-    .slice(0, 50)
-}
-
-function normalizeGrepParams(params, context) {
-  const root = resolveWorkspacePath(params?.path || '.', context)
-  const query = String(params?.query || '')
-  if (!query) {
-    const error = new Error('query is required')
-    error.statusCode = 400
-    throw error
-  }
-
-  const flags = params?.caseSensitive ? 'g' : 'gi'
-  try {
-    params?.regex
-      ? new RegExp(query, flags)
-      : new RegExp(escapeRegExp(query), flags)
-  } catch {
-    const error = new Error('Invalid regular expression')
-    error.statusCode = 400
-    throw error
-  }
-
-  return {
-    root,
-    query,
-    regex: Boolean(params?.regex),
-    caseSensitive: Boolean(params?.caseSensitive),
-    limit: clampNumber(params?.limit, 200, 1, 1000),
-    glob: normalizeGlobList(params?.glob),
-    context: clampNumber(params?.context, 0, 0, 20),
-    beforeContext: clampNumber(params?.beforeContext, 0, 0, 20),
-    afterContext: clampNumber(params?.afterContext, 0, 0, 20),
-    filesWithMatches: Boolean(params?.filesWithMatches),
-    respectGitIgnore: Boolean(params?.respectGitIgnore),
-  }
-}
-
-function isRegexLikelyRipgrepCompatible(query) {
-  return !(/\(\?[=!<]/.test(query) || /\\[1-9]/.test(query))
-}
-
-function ripgrepCandidatePath() {
-  try {
-    return require('@vscode/ripgrep').rgPath || null
-  } catch {
-    return null
-  }
-}
-
-async function verifyRipgrepExecutable(command) {
-  return new Promise((resolve) => {
-    const child = spawn(command, ['--version'], { shell: false, windowsHide: true })
-    child.once('error', () => resolve(false))
-    child.once('close', (code) => resolve(code === 0))
-  })
-}
-
-async function resolveRipgrepExecutable() {
-  if (cachedRipgrepExecutable !== undefined) return cachedRipgrepExecutable
-
-  const bundled = ripgrepCandidatePath()
-  if (bundled && await verifyRipgrepExecutable(bundled)) {
-    cachedRipgrepExecutable = { command: bundled, source: 'bundled' }
-    return cachedRipgrepExecutable
-  }
-
-  if (await verifyRipgrepExecutable('rg')) {
-    cachedRipgrepExecutable = { command: 'rg', source: 'system' }
-    return cachedRipgrepExecutable
-  }
-
-  cachedRipgrepExecutable = null
-  return cachedRipgrepExecutable
-}
-
-function buildRipgrepArgs(options, context) {
-  const args = [
-    '--line-number',
-    '--color=never',
-    '--max-filesize',
-    RIPGREP_MAX_FILESIZE,
-  ]
-
-  if (options.filesWithMatches) {
-    args.push('--files-with-matches')
-  } else {
-    args.push('--json')
-  }
-  if (!options.regex) args.push('--fixed-strings')
-  if (!options.caseSensitive) args.push('--ignore-case')
-  if (!options.respectGitIgnore) args.push('--hidden', '--no-ignore')
-  if (options.context > 0) args.push('-C', String(options.context))
-  if (options.beforeContext > 0) args.push('-B', String(options.beforeContext))
-  if (options.afterContext > 0) args.push('-A', String(options.afterContext))
-
-  for (const pattern of options.glob) args.push('--glob', pattern)
-  for (const pattern of DEFAULT_EXCLUDE_GLOBS) args.push('--glob', pattern)
-  for (const pattern of SENSITIVE_EXCLUDE_GLOBS) args.push('--glob', pattern)
-
-  args.push('--', options.query, toWorkspaceRelative(options.root, context) || '.')
-  return args
-}
-
-function cleanRipgrepLine(value) {
-  return String(value || '').replace(/[\r\n]+$/, '')
-}
-
-function ripgrepRelativePath(value) {
-  return String(value || '').replace(/\\/g, '/')
-}
-
-function formatRipgrepJsonEvent(event) {
-  if (event?.type !== 'match' && event?.type !== 'context') return null
-  const data = event.data || {}
-  const file = ripgrepRelativePath(data.path?.text)
-  const lineNumber = data.line_number
-  const line = cleanRipgrepLine(data.lines?.text)
-  if (!file || !lineNumber) return null
-  const separator = event.type === 'match' ? ':' : '-'
-  return `${file}:${lineNumber}${separator} ${line}`
-}
-
-function fallbackDetails(extra = {}) {
-  return Object.fromEntries(Object.entries(extra).filter(([, value]) => value !== undefined && value !== null && value !== ''))
-}
-
-async function grepFilesWithRipgrep(executable, options, context, runtime = {}) {
-  if (options.regex && !isRegexLikelyRipgrepCompatible(options.query)) {
-    throw new Error('regex uses JavaScript-only features that ripgrep does not support')
-  }
-
-  const cwd = getToolWorkspaceRoot(context)
-  const args = buildRipgrepArgs(options, context)
-  const matches = []
-  let stderr = ''
-  let buffer = ''
-  let killedForLimit = false
-  let settled = false
-
-  await new Promise((resolve, reject) => {
-    if (runtime.signal?.aborted) {
-      reject(new Error('Search aborted'))
-      return
-    }
-
-    const child = spawn(executable.command, args, {
-      cwd,
-      shell: false,
-      stdio: ['ignore', 'pipe', 'pipe'],
-      windowsHide: true,
-    })
-
-    const cleanup = () => {
-      clearTimeout(timer)
-      runtime.signal?.removeEventListener?.('abort', onAbort)
-    }
-    const finish = (error) => {
-      if (settled) return
-      settled = true
-      cleanup()
-      if (error) reject(error)
-      else resolve()
-    }
-    const stopForLimit = () => {
-      if (killedForLimit) return
-      killedForLimit = true
-      killProcessTree(child, 'SIGTERM')
-    }
-    const processLine = (line) => {
-      if (!line || matches.length >= options.limit) return
-      if (options.filesWithMatches) {
-        matches.push(ripgrepRelativePath(line))
-      } else {
-        try {
-          const formatted = formatRipgrepJsonEvent(JSON.parse(line))
-          if (formatted) matches.push(formatted)
-        } catch {
-          // Ignore malformed partial output and let process exit handling decide fallback.
-        }
-      }
-      if (matches.length >= options.limit) stopForLimit()
-    }
-    const flushLines = (chunk) => {
-      buffer += chunk.toString()
-      const lines = buffer.split(/\r?\n/)
-      buffer = lines.pop() || ''
-      for (const line of lines) processLine(line)
-    }
-    function onAbort() {
-      killProcessTree(child, 'SIGTERM')
-      finish(new Error('Search aborted'))
-    }
-
-    const timer = setTimeout(() => {
-      killProcessTree(child, 'SIGTERM')
-      finish(new Error('ripgrep search timed out'))
-    }, RIPGREP_TIMEOUT_MS)
-
-    runtime.signal?.addEventListener?.('abort', onAbort, { once: true })
-    child.stdout.on('data', flushLines)
-    child.stderr.on('data', (chunk) => {
-      stderr = truncateText(stderr + chunk.toString(), 2000)
-    })
-    child.once('error', finish)
-    child.once('close', (code) => {
-      if (buffer) processLine(buffer)
-      if (killedForLimit || code === 0 || code === 1) {
-        finish()
-      } else {
-        finish(new Error(stderr.trim() || `ripgrep exited with code ${code}`))
-      }
-    })
-  })
-
-  return {
-    content: matches.length ? truncateText(matches.slice(0, options.limit).join('\n')) : 'No matches found.',
-    details: {
-      path: toWorkspaceRelative(options.root, context),
-      project: context?.project,
-      query: options.query,
-      count: matches.length,
-      limit: options.limit,
-      backend: 'ripgrep',
-      ripgrepSource: executable.source,
-    },
-  }
-}
-
-async function grepFilesWithNode(options, context, extraDetails = {}) {
-  const flags = options.caseSensitive ? 'g' : 'gi'
-  const matcher = options.regex
-    ? new RegExp(options.query, flags)
-    : new RegExp(escapeRegExp(options.query), flags)
-
-  const files = await walkFiles(options.root, [], context)
-  const matches = []
-
-  // Stat and filter files in parallel, then grep in parallel batches.
-  const candidateResults = await poolMap(files, async (file) => {
-    try {
-      const stat = await fs.stat(file)
-      if (stat.size > 1024 * 1024) return { file, skip: true }
-      return { file, skip: false }
-    } catch {
-      return { file, skip: true }
-    }
-  })
-
-  const candidates = candidateResults.filter((r) => !r.skip).map((r) => r.file)
-
-  // Grep with bounded concurrency — short-circuit when limit reached.
-  let matchCount = 0
-  const filesWithMatches = new Set()
-  for (let batchStart = 0; batchStart < candidates.length && matchCount < options.limit; batchStart += 20) {
-    const batch = candidates.slice(batchStart, batchStart + 20)
-    const batchMatches = await Promise.all(
-      batch.map(async (file) => {
-        if (matchCount >= options.limit) return []
-        try {
-          const text = await fs.readFile(file, 'utf8')
-          const lines = splitLines(text)
-          const fileMatches = []
-          for (let index = 0; index < lines.length && (matchCount + fileMatches.length) < options.limit; index++) {
-            matcher.lastIndex = 0
-            if (matcher.test(lines[index])) {
-              const relative = toWorkspaceRelative(file, context)
-              if (options.filesWithMatches) {
-                if (!filesWithMatches.has(relative)) {
-                  filesWithMatches.add(relative)
-                  fileMatches.push(relative)
-                }
-              } else {
-                fileMatches.push(`${relative}:${index + 1}: ${lines[index]}`)
-              }
-            }
-          }
-          return fileMatches
-        } catch {
-          return []
-        }
-      }),
-    )
-    for (const fm of batchMatches) {
-      if (matchCount >= options.limit) break
-      for (const m of fm) {
-        if (matchCount >= options.limit) break
-        matches.push(m)
-        matchCount++
-      }
-    }
-  }
-
-  return {
-    content: matches.length ? truncateText(matches.join('\n')) : 'No matches found.',
-    details: {
-      path: toWorkspaceRelative(options.root, context),
-      project: context?.project,
-      query: options.query,
-      count: matches.length,
-      limit: options.limit,
-      backend: 'node',
-      ...fallbackDetails(extraDetails),
-    },
-  }
-}
-
-export async function toolGrepFiles(params, context, runtime = {}) {
-  const options = normalizeGrepParams(params, context)
-  await assertSafeWorkspacePath(options.root, context)
-
-  const executable = await resolveRipgrepExecutable()
-  if (executable) {
-    try {
-      return await grepFilesWithRipgrep(executable, options, context, runtime)
-    } catch (error) {
-      if (runtime.signal?.aborted) throw error
-      return grepFilesWithNode(options, context, {
-        fallbackFrom: 'ripgrep',
-        fallbackReason: error?.message || 'ripgrep unavailable',
-      })
-    }
-  }
-
-  return grepFilesWithNode(options, context, { fallbackReason: 'ripgrep unavailable' })
-}
-
-// --- write_file ---
-export async function toolWriteFile(params, context) {
-  const file = resolveWorkspacePath(params?.path, context)
-  await assertSafeWorkspacePath(file, context, { forWrite: true })
-
-  const content = String(params?.content ?? '')
-  const relativePath = toWorkspaceRelative(file, context)
-  let oldText = ''
-  let existed = true
-  try {
-    oldText = await fs.readFile(file, 'utf8')
-  } catch (error) {
-    if (error?.code !== 'ENOENT') throw error
-    existed = false
-  }
-  const diff = createTextDiff(oldText, content, relativePath, { oldExists: existed })
-
-  await fs.mkdir(path.dirname(file), { recursive: true })
-  await fs.writeFile(file, content, 'utf8')
-
-  return {
-    content: `${existed ? 'Wrote' : 'Created'} ${relativePath} (+${diff.addedLines} -${diff.removedLines})`,
-    details: { path: relativePath, project: context?.project, bytes: Buffer.byteLength(content, 'utf8'), created: !existed, diff },
-  }
-}
-
-// --- edit_file ---
-function countOccurrences(text, needle) {
-  if (!needle) return 0
-  let count = 0
-  let index = 0
-  while ((index = text.indexOf(needle, index)) !== -1) {
-    count++
-    index += needle.length
-  }
-  return count
-}
-
-function detectLineEnding(text) {
-  return text.includes('\r\n') ? '\r\n' : '\n'
-}
-
-function normalizeLineEndings(text) {
-  return text.replaceAll('\r\n', '\n').replaceAll('\r', '\n')
-}
-
-function convertToLineEnding(text, ending) {
-  const normalized = normalizeLineEndings(text)
-  return ending === '\r\n' ? normalized.replaceAll('\n', '\r\n') : normalized
-}
-
-export async function toolEditFile(params, context) {
-  const file = resolveWorkspacePath(params?.path, context)
-  await assertSafeWorkspacePath(file, context)
-
-  const rawOldText = String(params?.oldText ?? '')
-  const rawNewText = String(params?.newText ?? '')
-  const text = await fs.readFile(file, 'utf8')
-  const lineEnding = detectLineEnding(text)
-  const oldText = convertToLineEnding(rawOldText, lineEnding)
-  const newText = convertToLineEnding(rawNewText, lineEnding)
-  const count = countOccurrences(text, oldText)
-
-  if (count !== 1) {
-    const rawCount = oldText === rawOldText ? count : countOccurrences(text, rawOldText)
-    const suffix = rawCount !== count ? ` after normalizing line endings; raw match count was ${rawCount}` : ''
-    const error = new Error(`oldText must match exactly once; found ${count} matches${suffix}`)
-    error.statusCode = 400
-    throw error
-  }
-
-  const nextText = text.replace(oldText, newText)
-  const relativePath = toWorkspaceRelative(file, context)
-  const diff = createTextDiff(text, nextText, relativePath)
-
-  await fs.writeFile(file, nextText, 'utf8')
-
-  return {
-    content: `Edited ${relativePath} (+${diff.addedLines} -${diff.removedLines})`,
-    details: { path: relativePath, project: context?.project, replaced: count, diff },
-  }
-}
-
-// --- run_command ---
-function activeSkillsForContext(context) {
-  return mergeSkills(context?.globalSkills, context?.projectSkills)
-}
-
-function activeSkillByName(context, name) {
-  const skillName = String(name || '')
-  return activeSkillsForContext(context).find((skill) => skill.name === skillName)
-}
-
-export async function loadSkillToolContext(config = {}) {
-  const globalSkills = await loadSelectedGlobalSkills(config.globalSkillNames)
-  const projectSkills = config.workspaceRoot
-    ? await loadSelectedProjectSkills(config.projectSkillNames, config.workspaceRoot)
-    : []
-  return { globalSkills, projectSkills }
-}
-
-// --- activate_skill ---
-export async function toolActivateSkill(params, context) {
-  const skill = activeSkillByName(context, params?.name)
-  if (!skill) {
-    const error = new Error(`Unknown or disabled skill: ${params?.name || ''}`)
-    error.statusCode = 404
-    throw error
-  }
-
-  return {
-    content: truncateText(await formatSkillActivation(skill)),
-    details: {
-      skill: skill.name,
-      source: skill.source,
-      directory: skill.rootDir,
-    },
-  }
-}
-
-// --- read_skill_resource ---
-export async function toolReadSkillResource(params, context) {
-  const skill = activeSkillByName(context, params?.skill)
-  if (!skill) {
-    const error = new Error(`Unknown or disabled skill: ${params?.skill || ''}`)
-    error.statusCode = 404
-    throw error
-  }
-
-  const result = await readSkillResource(skill, params?.path, params)
-  return {
-    content: truncateText(result.content),
-    details: result.details,
-  }
-}
-
-// --- run_command ---
-const DEFAULT_RUN_COMMAND_TIMEOUT_MS = 30 * 60 * 1000
-const MIN_RUN_COMMAND_TIMEOUT_MS = 1000
-const MAX_RUN_COMMAND_TIMEOUT_MS = 30 * 60 * 1000
-const COMMAND_TAIL_LINES = 100
-
-function formatDurationMs(durationMs = 0) {
-  const totalSeconds = Math.max(0, Math.floor(durationMs / 1000))
-  const minutes = Math.floor(totalSeconds / 60)
-  const seconds = totalSeconds % 60
-  if (minutes > 0) return `${minutes}m ${seconds}s (${durationMs}ms)`
-  return `${seconds}s (${durationMs}ms)`
-}
-
-function tailText(current, chunk, maxLines = COMMAND_TAIL_LINES) {
-  const lines = (current + chunk).split(/\r?\n/)
-  if (lines.length <= maxLines) return lines.join('\n')
-  return lines.slice(lines.length - maxLines).join('\n')
-}
-
-function tailLabel(name, truncated) {
-  return truncated ? `${name} (last ${COMMAND_TAIL_LINES} lines):` : `${name}:`
-}
-
-function commandStatus(meta = {}) {
-  if (meta.running) return 'Status: running'
-  const flags = [
-    meta.timedOut ? 'timed out' : null,
-    meta.aborted ? 'aborted' : null,
-  ].filter(Boolean)
-  const suffix = flags.length ? ` (${flags.join(', ')})` : ''
-  return `Exit code: ${meta.code ?? 'unknown'}${meta.signal ? `, signal: ${meta.signal}` : ''}${suffix}`
-}
-
-function formatCommandOutput(command, stdout, stderr, meta = {}) {
-  const lines = [
-    `Command: ${command}`,
-  ]
-  if (meta.description) lines.push(`Description: ${meta.description}`)
-  lines.push(commandStatus(meta))
-  if (typeof meta.durationMs === 'number') lines.push(`Duration: ${formatDurationMs(meta.durationMs)}`)
-  if (typeof meta.timeoutMs === 'number') lines.push(`Timeout: ${formatDurationMs(meta.timeoutMs)}`)
-  if (meta.cwd) lines.push(`CWD: ${meta.cwd}`)
-  if (meta.outputFile) lines.push(`Full output: ${meta.outputFile}`)
-  if (meta.outputTruncated) lines.push(`Output mode: showing the last ${COMMAND_TAIL_LINES} lines; full output is saved to the log file.`)
-  if (meta.logError) lines.push(`Log warning: ${meta.logError}`)
-  lines.push('', tailLabel('STDOUT', meta.stdoutTruncated), stdout || '(empty)', '', tailLabel('STDERR', meta.stderrTruncated), stderr || '(empty)')
-  return lines.join('\n')
-}
-
-function safeLogFilePart(value, fallback) {
-  const text = String(value || '').replace(/[^a-zA-Z0-9_.-]/g, '_').slice(0, 80)
-  return text || fallback
-}
-
-async function createCommandLogStream(command, { cwd, description, toolCallId } = {}) {
-  const dir = path.join(logsDir, 'commands')
-  await fs.mkdir(dir, { recursive: true })
-  const timestamp = new Date().toISOString().replace(/[:.]/g, '-')
-  const fileName = `${timestamp}_${safeLogFilePart(toolCallId, 'command')}.log`
-  const outputFile = path.join(dir, fileName)
-  const stream = createWriteStream(outputFile, { flags: 'wx' })
-  stream.write([
-    `Command: ${command}`,
-    description ? `Description: ${description}` : null,
-    `CWD: ${cwd}`,
-    `Started at: ${new Date().toISOString()}`,
-    '',
-  ].filter(Boolean).join('\n'))
-  return { stream, outputFile }
-}
-
-function writeCommandLog(stream, source, chunk) {
-  stream.write(`\n[${source} ${new Date().toISOString()}]\n`)
-  stream.write(chunk)
-}
-
-function killProcessTree(child, signal = 'SIGTERM') {
-  if (!child?.pid) return
-
-  if (process.platform === 'win32') {
-    const killer = spawn('taskkill', ['/pid', String(child.pid), '/T', '/F'], {
-      stdio: 'ignore',
-      windowsHide: true,
-    })
-    killer.on('error', () => {
-      try { child.kill(signal) } catch { /* ignore */ }
-    })
-    return
-  }
-
-  try {
-    process.kill(-child.pid, signal)
-  } catch {
-    try { child.kill(signal) } catch { /* ignore */ }
-  }
-}
-
-const runningCommands = new Map()
-
-export function abortRunningCommand(toolCallId) {
-  if (!toolCallId) return false
-  const stop = runningCommands.get(toolCallId)
-  if (!stop) return false
-  stop('abort')
-  return true
-}
-
-export async function toolRunCommand(params, context, runtime = {}) {
-  const command = String(params?.command || '')
-  if (!command.trim()) {
-    const error = new Error('command is required')
-    error.statusCode = 400
-    throw error
-  }
-
-  const description = String(params?.description || '').trim().slice(0, 500)
-  const timeoutMs = clampNumber(params?.timeoutMs, DEFAULT_RUN_COMMAND_TIMEOUT_MS, MIN_RUN_COMMAND_TIMEOUT_MS, MAX_RUN_COMMAND_TIMEOUT_MS)
-  const cwd = getToolWorkspaceRoot(context)
-  const startedAt = Date.now()
-
-  if (runtime.signal?.aborted) {
-    const content = formatCommandOutput(command, '', 'Command aborted before start.', {
-      aborted: true,
-      cwd,
-      description,
-      timeoutMs,
-      durationMs: 0,
-    })
-    return { content: truncateText(content), details: { command, description, project: context?.project, cwd, timeoutMs, durationMs: 0, aborted: true } }
-  }
-
-  let logStream = null
-  let outputFile = null
-  let logError = null
-  try {
-    const log = await createCommandLogStream(command, { cwd, description, toolCallId: runtime.toolCallId })
-    logStream = log.stream
-    outputFile = log.outputFile
-    logStream.on('error', (error) => {
-      logError = error?.message || 'Failed to write command log.'
-    })
-  } catch (error) {
-    logError = error?.message || 'Failed to create command log.'
-  }
-
-  return new Promise((resolve) => {
-    const child = spawn(command, {
-      cwd,
-      shell: true,
-      stdio: ['ignore', 'pipe', 'pipe'],
-      windowsHide: true,
-      detached: process.platform !== 'win32',
-    })
-
-    let stdout = ''
-    let stderr = ''
-    let stdoutTruncated = false
-    let stderrTruncated = false
-    let timedOut = false
-    let aborted = false
-    let settled = false
-    let updateTimer = null
-    let updatePending = false
-    let forceKillTimer = null
-
-    const cleanup = () => {
-      clearTimeout(timer)
-      if (forceKillTimer) clearTimeout(forceKillTimer)
-      if (updateTimer) clearTimeout(updateTimer)
-      if (runtime.toolCallId) runningCommands.delete(runtime.toolCallId)
-      runtime.signal?.removeEventListener?.('abort', onAbort)
-    }
-
-    const commonDetails = (extra = {}) => {
-      const now = Date.now()
-      return {
-        command,
-        description,
-        project: context?.project,
-        cwd,
-        timeoutMs,
-        outputFile,
-        stdout,
-        stderr,
-        stdoutTruncated,
-        stderrTruncated,
-        outputTruncated: stdoutTruncated || stderrTruncated,
-        durationMs: now - startedAt,
-        toolCallId: runtime.toolCallId,
-        logError,
-        ...extra,
-      }
-    }
-
-    const resolveAfterLogClose = (result) => {
-      if (!logStream) {
-        resolve(result)
-        return
-      }
-      const details = result.details || {}
-      logStream.write([
-        '',
-        '',
-        `[quickforge ${new Date().toISOString()}]`,
-        `Exit code: ${details.code ?? 'unknown'}${details.signal ? `, signal: ${details.signal}` : ''}`,
-        `Duration: ${formatDurationMs(details.durationMs)}`,
-        `Timed out: ${Boolean(details.timedOut)}`,
-        `Aborted: ${Boolean(details.aborted)}`,
-        'Command finished.',
-        '',
-      ].join('\n'))
-      logStream.end(() => resolve(result))
-    }
-
-    const finish = ({ code = null, signal = null, error = null } = {}) => {
-      if (settled) return
-      flushUpdate()
-      settled = true
-      cleanup()
-      const durationMs = Date.now() - startedAt
-      if (error) {
-        const details = commonDetails({ error: error.message, aborted, timedOut, durationMs })
-        resolveAfterLogClose({
-          isError: true,
-          content: truncateText(formatCommandOutput(command, stdout, `Error running command: ${error.message}\n${stderr}`.trim(), details)),
-          details,
-        })
-        return
-      }
-      const details = commonDetails({ code, signal, timedOut, aborted, durationMs })
-      const content = formatCommandOutput(command, stdout, stderr, details)
-      resolveAfterLogClose({ content: truncateText(content), details })
-    }
-
-    const stopChild = (reason) => {
-      if (reason === 'timeout') timedOut = true
-      if (reason === 'abort') aborted = true
-      killProcessTree(child, 'SIGTERM')
-      forceKillTimer = setTimeout(() => {
-        killProcessTree(child, 'SIGKILL')
-      }, 1500)
-    }
-
-    if (runtime.toolCallId) runningCommands.set(runtime.toolCallId, stopChild)
-
-    function onAbort() {
-      stopChild('abort')
-      finish({ signal: 'SIGTERM' })
-    }
-
-    const runningDetails = () => commonDetails({ running: true })
-
-    const emitUpdate = () => {
-      updateTimer = null
-      if (settled || !updatePending) return
-      updatePending = false
-      const details = runningDetails()
-      runtime.onUpdate?.({
-        content: [{ type: 'text', text: truncateText(formatCommandOutput(command, stdout, stderr, details)) }],
-        details,
-      })
-    }
-    const flushUpdate = () => {
-      if (updateTimer) {
-        clearTimeout(updateTimer)
-        updateTimer = null
-      }
-      if (!updatePending) return
-      updatePending = false
-      const details = runningDetails()
-      runtime.onUpdate?.({
-        content: [{ type: 'text', text: truncateText(formatCommandOutput(command, stdout, stderr, details)) }],
-        details,
-      })
-    }
-    const scheduleUpdate = () => {
-      if (settled) return
-      updatePending = true
-      if (!updateTimer) updateTimer = setTimeout(emitUpdate, 150)
-    }
-    const timer = setTimeout(() => {
-      stopChild('timeout')
-      finish({ signal: 'SIGTERM' })
-    }, timeoutMs)
-
-    runtime.signal?.addEventListener?.('abort', onAbort, { once: true })
-
-    child.stdout.on('data', (chunk) => {
-      if (settled) return
-      const text = chunk.toString()
-      stdoutTruncated = stdoutTruncated || (stdout + text).split(/\r?\n/).length > COMMAND_TAIL_LINES
-      stdout = tailText(stdout, text)
-      if (logStream && !logError) writeCommandLog(logStream, 'stdout', text)
-      scheduleUpdate()
-    })
-    child.stderr.on('data', (chunk) => {
-      if (settled) return
-      const text = chunk.toString()
-      stderrTruncated = stderrTruncated || (stderr + text).split(/\r?\n/).length > COMMAND_TAIL_LINES
-      stderr = tailText(stderr, text)
-      if (logStream && !logError) writeCommandLog(logStream, 'stderr', text)
-      scheduleUpdate()
-    })
-    child.on('close', (code, signal) => {
-      finish({ code, signal })
-    })
-    child.on('error', (err) => {
-      finish({ error: err })
-    })
-  })
-}
-
-export const toolHandlers = {
-  read_file: toolReadFile,
-  grep_files: toolGrepFiles,
-  write_file: toolWriteFile,
-  edit_file: toolEditFile,
-  run_command: toolRunCommand,
-  activate_skill: toolActivateSkill,
-  read_skill_resource: toolReadSkillResource,
-}
+import { createWriteStream, promises as fs } from 'node:fs'
+import path from 'node:path'
+import { spawn } from 'node:child_process'
+import { createRequire } from 'node:module'
+import { resolveWorkspacePath, toWorkspaceRelative, assertSafeWorkspacePath, truncateText, splitLines, walkFiles } from '../utils/workspace.mjs'
+import { logsDir } from '../storage.mjs'
+import { createTextDiff } from '../utils/text-diff.mjs'
+import {
+  formatSkillActivation,
+  loadSelectedGlobalSkills,
+  loadSelectedProjectSkills,
+  mergeSkills,
+  readSkillResource,
+} from '../skills.mjs'
+import { getToolWorkspaceRoot } from '../utils/workspace.mjs'
+
+const require = createRequire(import.meta.url)
+
+// --- read_file ---
+export async function toolReadFile(params, context) {
+  const file = resolveWorkspacePath(params?.path, context)
+  await assertSafeWorkspacePath(file, context)
+
+  const text = await fs.readFile(file, 'utf8')
+  const lines = splitLines(text)
+  const offset = Math.max(1, Number(params?.offset || 1))
+  const limit = Math.min(2000, Math.max(1, Number(params?.limit || 200)))
+  const selected = lines.slice(offset - 1, offset - 1 + limit)
+  const content = selected.map((line, index) => `${offset + index}: ${line}`).join('\n')
+  const suffix = offset - 1 + limit < lines.length ? `\n\n[showing ${selected.length} of ${lines.length} lines]` : ''
+
+  return {
+    content: truncateText(`${content}${suffix}`),
+    details: { path: toWorkspaceRelative(file, context), project: context?.project, totalLines: lines.length, offset, limit },
+  }
+}
+
+// --- grep_files ---
+
+const RIPGREP_MAX_FILESIZE = '1M'
+const RIPGREP_TIMEOUT_MS = 60 * 1000
+const DEFAULT_EXCLUDE_GLOBS = [
+  '!.git/**',
+  '!node_modules/**',
+  '!dist/**',
+  '!dist-ssr/**',
+  '!.vite/**',
+  '!**/*.png',
+  '!**/*.jpg',
+  '!**/*.jpeg',
+  '!**/*.gif',
+  '!**/*.webp',
+  '!**/*.ico',
+  '!**/*.pdf',
+  '!**/*.zip',
+  '!**/*.gz',
+  '!**/*.7z',
+  '!**/*.exe',
+  '!**/*.dll',
+  '!**/*.woff',
+  '!**/*.woff2',
+  '!**/*.ttf',
+]
+const SENSITIVE_EXCLUDE_GLOBS = [
+  '!.env',
+  '!**/.env',
+  '!.env.*',
+  '!**/.env.*',
+  '!**/*.pem',
+  '!**/*.key',
+  '!**/*.p12',
+  '!**/*.pfx',
+  '!**/*.crt',
+  '!**/*.cer',
+  '!**/*.token',
+  '!credentials.json',
+  '!**/credentials.json',
+  '!secrets.json',
+  '!**/secrets.json',
+  '!id_rsa',
+  '!**/id_rsa',
+  '!id_ed25519',
+  '!**/id_ed25519',
+]
+
+let cachedRipgrepExecutable
+
+/**
+ * Process items with bounded concurrency.  Returns results in input order.
+ * @template T, R
+ * @param {T[]} items
+ * @param {(item: T, index: number) => Promise<R>} fn
+ * @param {number} concurrency
+ * @returns {Promise<R[]>}
+ */
+async function poolMap(items, fn, concurrency = 20) {
+  const results = new Array(items.length)
+  let cursor = 0
+
+  async function worker() {
+    while (cursor < items.length) {
+      const index = cursor++
+      results[index] = await fn(items[index], index)
+    }
+  }
+
+  const workers = Array.from({ length: Math.min(concurrency, items.length) }, () => worker())
+  await Promise.all(workers)
+  return results
+}
+
+function clampNumber(value, defaultValue, min, max) {
+  const number = Number(value)
+  if (!Number.isFinite(number)) return defaultValue
+  return Math.min(max, Math.max(min, Math.trunc(number)))
+}
+
+function escapeRegExp(value) {
+  return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')
+}
+
+function normalizeGlobList(value) {
+  const values = Array.isArray(value) ? value : typeof value === 'string' ? [value] : []
+  return values
+    .map((item) => String(item || '').trim())
+    .filter(Boolean)
+    .slice(0, 50)
+}
+
+function normalizeGrepParams(params, context) {
+  const root = resolveWorkspacePath(params?.path || '.', context)
+  const query = String(params?.query || '')
+  if (!query) {
+    const error = new Error('query is required')
+    error.statusCode = 400
+    throw error
+  }
+
+  const flags = params?.caseSensitive ? 'g' : 'gi'
+  try {
+    params?.regex
+      ? new RegExp(query, flags)
+      : new RegExp(escapeRegExp(query), flags)
+  } catch {
+    const error = new Error('Invalid regular expression')
+    error.statusCode = 400
+    throw error
+  }
+
+  return {
+    root,
+    query,
+    regex: Boolean(params?.regex),
+    caseSensitive: Boolean(params?.caseSensitive),
+    limit: clampNumber(params?.limit, 200, 1, 1000),
+    glob: normalizeGlobList(params?.glob),
+    context: clampNumber(params?.context, 0, 0, 20),
+    beforeContext: clampNumber(params?.beforeContext, 0, 0, 20),
+    afterContext: clampNumber(params?.afterContext, 0, 0, 20),
+    filesWithMatches: Boolean(params?.filesWithMatches),
+    respectGitIgnore: Boolean(params?.respectGitIgnore),
+  }
+}
+
+function isRegexLikelyRipgrepCompatible(query) {
+  return !(/\(\?[=!<]/.test(query) || /\\[1-9]/.test(query))
+}
+
+function ripgrepCandidatePath() {
+  try {
+    return require('@vscode/ripgrep').rgPath || null
+  } catch {
+    return null
+  }
+}
+
+async function verifyRipgrepExecutable(command) {
+  return new Promise((resolve) => {
+    const child = spawn(command, ['--version'], { shell: false, windowsHide: true })
+    child.once('error', () => resolve(false))
+    child.once('close', (code) => resolve(code === 0))
+  })
+}
+
+async function resolveRipgrepExecutable() {
+  if (cachedRipgrepExecutable !== undefined) return cachedRipgrepExecutable
+
+  const bundled = ripgrepCandidatePath()
+  if (bundled && await verifyRipgrepExecutable(bundled)) {
+    cachedRipgrepExecutable = { command: bundled, source: 'bundled' }
+    return cachedRipgrepExecutable
+  }
+
+  if (await verifyRipgrepExecutable('rg')) {
+    cachedRipgrepExecutable = { command: 'rg', source: 'system' }
+    return cachedRipgrepExecutable
+  }
+
+  cachedRipgrepExecutable = null
+  return cachedRipgrepExecutable
+}
+
+function buildRipgrepArgs(options, context) {
+  const args = [
+    '--line-number',
+    '--color=never',
+    '--max-filesize',
+    RIPGREP_MAX_FILESIZE,
+  ]
+
+  if (options.filesWithMatches) {
+    args.push('--files-with-matches')
+  } else {
+    args.push('--json')
+  }
+  if (!options.regex) args.push('--fixed-strings')
+  if (!options.caseSensitive) args.push('--ignore-case')
+  if (!options.respectGitIgnore) args.push('--hidden', '--no-ignore')
+  if (options.context > 0) args.push('-C', String(options.context))
+  if (options.beforeContext > 0) args.push('-B', String(options.beforeContext))
+  if (options.afterContext > 0) args.push('-A', String(options.afterContext))
+
+  for (const pattern of options.glob) args.push('--glob', pattern)
+  for (const pattern of DEFAULT_EXCLUDE_GLOBS) args.push('--glob', pattern)
+  for (const pattern of SENSITIVE_EXCLUDE_GLOBS) args.push('--glob', pattern)
+
+  args.push('--', options.query, toWorkspaceRelative(options.root, context) || '.')
+  return args
+}
+
+function cleanRipgrepLine(value) {
+  return String(value || '').replace(/[\r\n]+$/, '')
+}
+
+function ripgrepRelativePath(value) {
+  return String(value || '').replace(/\\/g, '/')
+}
+
+function formatCommandArg(value) {
+  const text = String(value ?? '')
+  return /^[A-Za-z0-9_@%+=:,./\\-]+$/.test(text) ? text : JSON.stringify(text)
+}
+
+function formatSpawnCommand(command, args) {
+  return [command, ...args].map(formatCommandArg).join(' ')
+}
+
+function grepSearchOptionsDetails(options) {
+  return {
+    regex: options.regex,
+    caseSensitive: options.caseSensitive,
+    glob: options.glob,
+    context: options.context,
+    beforeContext: options.beforeContext,
+    afterContext: options.afterContext,
+    filesWithMatches: options.filesWithMatches,
+    respectGitIgnore: options.respectGitIgnore,
+    maxFileSize: RIPGREP_MAX_FILESIZE,
+    defaultExcludeGlobs: DEFAULT_EXCLUDE_GLOBS,
+    sensitiveExcludeGlobs: SENSITIVE_EXCLUDE_GLOBS,
+  }
+}
+
+function formatRipgrepJsonEvent(event) {
+  if (event?.type !== 'match' && event?.type !== 'context') return null
+  const data = event.data || {}
+  const file = ripgrepRelativePath(data.path?.text)
+  const lineNumber = data.line_number
+  const line = cleanRipgrepLine(data.lines?.text)
+  if (!file || !lineNumber) return null
+  const separator = event.type === 'match' ? ':' : '-'
+  return `${file}:${lineNumber}${separator} ${line}`
+}
+
+function fallbackDetails(extra = {}) {
+  return Object.fromEntries(Object.entries(extra).filter(([, value]) => value !== undefined && value !== null && value !== ''))
+}
+
+async function grepFilesWithRipgrep(executable, options, context, runtime = {}) {
+  if (options.regex && !isRegexLikelyRipgrepCompatible(options.query)) {
+    throw new Error('regex uses JavaScript-only features that ripgrep does not support')
+  }
+
+  const cwd = getToolWorkspaceRoot(context)
+  const args = buildRipgrepArgs(options, context)
+  const matches = []
+  let stderr = ''
+  let buffer = ''
+  let killedForLimit = false
+  let settled = false
+
+  await new Promise((resolve, reject) => {
+    if (runtime.signal?.aborted) {
+      reject(new Error('Search aborted'))
+      return
+    }
+
+    const child = spawn(executable.command, args, {
+      cwd,
+      shell: false,
+      stdio: ['ignore', 'pipe', 'pipe'],
+      windowsHide: true,
+    })
+
+    const cleanup = () => {
+      clearTimeout(timer)
+      runtime.signal?.removeEventListener?.('abort', onAbort)
+    }
+    const finish = (error) => {
+      if (settled) return
+      settled = true
+      cleanup()
+      if (error) reject(error)
+      else resolve()
+    }
+    const stopForLimit = () => {
+      if (killedForLimit) return
+      killedForLimit = true
+      killProcessTree(child, 'SIGTERM')
+    }
+    const processLine = (line) => {
+      if (!line || matches.length >= options.limit) return
+      if (options.filesWithMatches) {
+        matches.push(ripgrepRelativePath(line))
+      } else {
+        try {
+          const formatted = formatRipgrepJsonEvent(JSON.parse(line))
+          if (formatted) matches.push(formatted)
+        } catch {
+          // Ignore malformed partial output and let process exit handling decide fallback.
+        }
+      }
+      if (matches.length >= options.limit) stopForLimit()
+    }
+    const flushLines = (chunk) => {
+      buffer += chunk.toString()
+      const lines = buffer.split(/\r?\n/)
+      buffer = lines.pop() || ''
+      for (const line of lines) processLine(line)
+    }
+    function onAbort() {
+      killProcessTree(child, 'SIGTERM')
+      finish(new Error('Search aborted'))
+    }
+
+    const timer = setTimeout(() => {
+      killProcessTree(child, 'SIGTERM')
+      finish(new Error('ripgrep search timed out'))
+    }, RIPGREP_TIMEOUT_MS)
+
+    runtime.signal?.addEventListener?.('abort', onAbort, { once: true })
+    child.stdout.on('data', flushLines)
+    child.stderr.on('data', (chunk) => {
+      stderr = truncateText(stderr + chunk.toString(), 2000)
+    })
+    child.once('error', finish)
+    child.once('close', (code) => {
+      if (buffer) processLine(buffer)
+      if (killedForLimit || code === 0 || code === 1) {
+        finish()
+      } else {
+        finish(new Error(stderr.trim() || `ripgrep exited with code ${code}`))
+      }
+    })
+  })
+
+  return {
+    content: matches.length ? truncateText(matches.slice(0, options.limit).join('\n')) : 'No matches found.',
+    details: {
+      path: toWorkspaceRelative(options.root, context),
+      project: context?.project,
+      query: options.query,
+      count: matches.length,
+      limit: options.limit,
+      backend: 'ripgrep',
+      ripgrepSource: executable.source,
+      cwd,
+      command: formatSpawnCommand(executable.command, args),
+      executable: executable.command,
+      args,
+      searchOptions: grepSearchOptionsDetails(options),
+    },
+  }
+}
+
+async function grepFilesWithNode(options, context, extraDetails = {}) {
+  const flags = options.caseSensitive ? 'g' : 'gi'
+  const matcher = options.regex
+    ? new RegExp(options.query, flags)
+    : new RegExp(escapeRegExp(options.query), flags)
+
+  const files = await walkFiles(options.root, [], context)
+  const matches = []
+
+  // Stat and filter files in parallel, then grep in parallel batches.
+  const candidateResults = await poolMap(files, async (file) => {
+    try {
+      const stat = await fs.stat(file)
+      if (stat.size > 1024 * 1024) return { file, skip: true }
+      return { file, skip: false }
+    } catch {
+      return { file, skip: true }
+    }
+  })
+
+  const candidates = candidateResults.filter((r) => !r.skip).map((r) => r.file)
+
+  // Grep with bounded concurrency — short-circuit when limit reached.
+  let matchCount = 0
+  const filesWithMatches = new Set()
+  for (let batchStart = 0; batchStart < candidates.length && matchCount < options.limit; batchStart += 20) {
+    const batch = candidates.slice(batchStart, batchStart + 20)
+    const batchMatches = await Promise.all(
+      batch.map(async (file) => {
+        if (matchCount >= options.limit) return []
+        try {
+          const text = await fs.readFile(file, 'utf8')
+          const lines = splitLines(text)
+          const fileMatches = []
+          for (let index = 0; index < lines.length && (matchCount + fileMatches.length) < options.limit; index++) {
+            matcher.lastIndex = 0
+            if (matcher.test(lines[index])) {
+              const relative = toWorkspaceRelative(file, context)
+              if (options.filesWithMatches) {
+                if (!filesWithMatches.has(relative)) {
+                  filesWithMatches.add(relative)
+                  fileMatches.push(relative)
+                }
+              } else {
+                fileMatches.push(`${relative}:${index + 1}: ${lines[index]}`)
+              }
+            }
+          }
+          return fileMatches
+        } catch {
+          return []
+        }
+      }),
+    )
+    for (const fm of batchMatches) {
+      if (matchCount >= options.limit) break
+      for (const m of fm) {
+        if (matchCount >= options.limit) break
+        matches.push(m)
+        matchCount++
+      }
+    }
+  }
+
+  return {
+    content: matches.length ? truncateText(matches.join('\n')) : 'No matches found.',
+    details: {
+      path: toWorkspaceRelative(options.root, context),
+      project: context?.project,
+      query: options.query,
+      count: matches.length,
+      limit: options.limit,
+      backend: 'node',
+      searchOptions: grepSearchOptionsDetails(options),
+      ...fallbackDetails(extraDetails),
+    },
+  }
+}
+
+export async function toolGrepFiles(params, context, runtime = {}) {
+  const options = normalizeGrepParams(params, context)
+  await assertSafeWorkspacePath(options.root, context)
+
+  const executable = await resolveRipgrepExecutable()
+  if (executable) {
+    try {
+      return await grepFilesWithRipgrep(executable, options, context, runtime)
+    } catch (error) {
+      if (runtime.signal?.aborted) throw error
+      return grepFilesWithNode(options, context, {
+        fallbackFrom: 'ripgrep',
+        fallbackReason: error?.message || 'ripgrep unavailable',
+      })
+    }
+  }
+
+  return grepFilesWithNode(options, context, { fallbackReason: 'ripgrep unavailable' })
+}
+
+// --- write_file ---
+export async function toolWriteFile(params, context) {
+  const file = resolveWorkspacePath(params?.path, context)
+  await assertSafeWorkspacePath(file, context, { forWrite: true })
+
+  const content = String(params?.content ?? '')
+  const relativePath = toWorkspaceRelative(file, context)
+  let oldText = ''
+  let existed = true
+  try {
+    oldText = await fs.readFile(file, 'utf8')
+  } catch (error) {
+    if (error?.code !== 'ENOENT') throw error
+    existed = false
+  }
+  const diff = createTextDiff(oldText, content, relativePath, { oldExists: existed })
+
+  await fs.mkdir(path.dirname(file), { recursive: true })
+  await fs.writeFile(file, content, 'utf8')
+
+  return {
+    content: `${existed ? 'Wrote' : 'Created'} ${relativePath} (+${diff.addedLines} -${diff.removedLines})`,
+    details: { path: relativePath, project: context?.project, bytes: Buffer.byteLength(content, 'utf8'), created: !existed, diff },
+  }
+}
+
+// --- edit_file ---
+function countOccurrences(text, needle) {
+  if (!needle) return 0
+  let count = 0
+  let index = 0
+  while ((index = text.indexOf(needle, index)) !== -1) {
+    count++
+    index += needle.length
+  }
+  return count
+}
+
+function detectLineEnding(text) {
+  return text.includes('\r\n') ? '\r\n' : '\n'
+}
+
+function normalizeLineEndings(text) {
+  return text.replaceAll('\r\n', '\n').replaceAll('\r', '\n')
+}
+
+function convertToLineEnding(text, ending) {
+  const normalized = normalizeLineEndings(text)
+  return ending === '\r\n' ? normalized.replaceAll('\n', '\r\n') : normalized
+}
+
+export async function toolEditFile(params, context) {
+  const file = resolveWorkspacePath(params?.path, context)
+  await assertSafeWorkspacePath(file, context)
+
+  const rawOldText = String(params?.oldText ?? '')
+  const rawNewText = String(params?.newText ?? '')
+  const text = await fs.readFile(file, 'utf8')
+  const lineEnding = detectLineEnding(text)
+  const oldText = convertToLineEnding(rawOldText, lineEnding)
+  const newText = convertToLineEnding(rawNewText, lineEnding)
+  const count = countOccurrences(text, oldText)
+
+  if (count !== 1) {
+    const rawCount = oldText === rawOldText ? count : countOccurrences(text, rawOldText)
+    const suffix = rawCount !== count ? ` after normalizing line endings; raw match count was ${rawCount}` : ''
+    const error = new Error(`oldText must match exactly once; found ${count} matches${suffix}`)
+    error.statusCode = 400
+    throw error
+  }
+
+  const nextText = text.replace(oldText, newText)
+  const relativePath = toWorkspaceRelative(file, context)
+  const diff = createTextDiff(text, nextText, relativePath)
+
+  await fs.writeFile(file, nextText, 'utf8')
+
+  return {
+    content: `Edited ${relativePath} (+${diff.addedLines} -${diff.removedLines})`,
+    details: { path: relativePath, project: context?.project, replaced: count, diff },
+  }
+}
+
+// --- run_command ---
+function activeSkillsForContext(context) {
+  return mergeSkills(context?.globalSkills, context?.projectSkills)
+}
+
+function activeSkillByName(context, name) {
+  const skillName = String(name || '')
+  return activeSkillsForContext(context).find((skill) => skill.name === skillName)
+}
+
+export async function loadSkillToolContext(config = {}) {
+  const globalSkills = await loadSelectedGlobalSkills(config.globalSkillNames)
+  const projectSkills = config.workspaceRoot
+    ? await loadSelectedProjectSkills(config.projectSkillNames, config.workspaceRoot)
+    : []
+  return { globalSkills, projectSkills }
+}
+
+// --- activate_skill ---
+export async function toolActivateSkill(params, context) {
+  const skill = activeSkillByName(context, params?.name)
+  if (!skill) {
+    const error = new Error(`Unknown or disabled skill: ${params?.name || ''}`)
+    error.statusCode = 404
+    throw error
+  }
+
+  return {
+    content: truncateText(await formatSkillActivation(skill)),
+    details: {
+      skill: skill.name,
+      source: skill.source,
+      directory: skill.rootDir,
+    },
+  }
+}
+
+// --- read_skill_resource ---
+export async function toolReadSkillResource(params, context) {
+  const skill = activeSkillByName(context, params?.skill)
+  if (!skill) {
+    const error = new Error(`Unknown or disabled skill: ${params?.skill || ''}`)
+    error.statusCode = 404
+    throw error
+  }
+
+  const result = await readSkillResource(skill, params?.path, params)
+  return {
+    content: truncateText(result.content),
+    details: result.details,
+  }
+}
+
+// --- run_command ---
+const DEFAULT_RUN_COMMAND_TIMEOUT_MS = 30 * 60 * 1000
+const MIN_RUN_COMMAND_TIMEOUT_MS = 1000
+const MAX_RUN_COMMAND_TIMEOUT_MS = 30 * 60 * 1000
+const COMMAND_PREVIEW_LINES = 200
+const COMMAND_PREVIEW_TOTAL_CHARS = 10000
+
+function formatDurationMs(durationMs = 0) {
+  const totalSeconds = Math.max(0, Math.floor(durationMs / 1000))
+  const minutes = Math.floor(totalSeconds / 60)
+  const seconds = totalSeconds % 60
+  if (minutes > 0) return `${minutes}m ${seconds}s (${durationMs}ms)`
+  return `${seconds}s (${durationMs}ms)`
+}
+
+function tailText(current, chunk, maxLines = COMMAND_PREVIEW_LINES, maxChars = COMMAND_PREVIEW_TOTAL_CHARS) {
+  const lines = (current + chunk).split(/\r?\n/)
+  const truncatedByLines = lines.length > maxLines
+  const lineLimitedText = truncatedByLines ? lines.slice(lines.length - maxLines).join('\n') : lines.join('\n')
+  const charLimitedText = tailChars(lineLimitedText, maxChars)
+  return {
+    text: charLimitedText.text,
+    truncated: truncatedByLines || charLimitedText.truncated,
+  }
+}
+
+function tailChars(text, maxChars) {
+  if (text.length <= maxChars) return { text, truncated: false }
+  return { text: text.slice(text.length - maxChars), truncated: true }
+}
+
+function balanceCommandPreviews(stdout, stderr, maxChars = COMMAND_PREVIEW_TOTAL_CHARS) {
+  if (stdout.length + stderr.length <= maxChars) {
+    return { stdout, stderr, stdoutTruncated: false, stderrTruncated: false }
+  }
+
+  const half = Math.floor(maxChars / 2)
+  let stdoutBudget = Math.min(stdout.length, half)
+  let stderrBudget = Math.min(stderr.length, half)
+  const remaining = maxChars - stdoutBudget - stderrBudget
+
+  if (remaining > 0) {
+    const stdoutNeed = Math.max(0, stdout.length - stdoutBudget)
+    const stderrNeed = Math.max(0, stderr.length - stderrBudget)
+    if (stderrNeed >= stdoutNeed) stderrBudget += Math.min(remaining, stderrNeed)
+    else stdoutBudget += Math.min(remaining, stdoutNeed)
+  }
+
+  const nextStdout = tailChars(stdout, stdoutBudget)
+  const nextStderr = tailChars(stderr, stderrBudget)
+  return {
+    stdout: nextStdout.text,
+    stderr: nextStderr.text,
+    stdoutTruncated: nextStdout.truncated,
+    stderrTruncated: nextStderr.truncated,
+  }
+}
+
+function tailLabel(name, truncated) {
+  return truncated ? `${name} preview (last ${COMMAND_PREVIEW_LINES} lines, shared ${COMMAND_PREVIEW_TOTAL_CHARS} chars):` : `${name} preview:`
+}
+
+function commandStatus(meta = {}) {
+  if (meta.running) return 'Status: running'
+  const flags = [
+    meta.timedOut ? 'timed out' : null,
+    meta.aborted ? 'aborted' : null,
+  ].filter(Boolean)
+  const suffix = flags.length ? ` (${flags.join(', ')})` : ''
+  return `Exit code: ${meta.code ?? 'unknown'}${meta.signal ? `, signal: ${meta.signal}` : ''}${suffix}`
+}
+
+function formatCommandOutput(command, stdout, stderr, meta = {}) {
+  const lines = [
+    `Command: ${command}`,
+  ]
+  if (meta.description) lines.push(`Description: ${meta.description}`)
+  lines.push(commandStatus(meta))
+  if (typeof meta.durationMs === 'number') lines.push(`Duration: ${formatDurationMs(meta.durationMs)}`)
+  if (typeof meta.timeoutMs === 'number') lines.push(`Timeout: ${formatDurationMs(meta.timeoutMs)}`)
+  if (meta.cwd) lines.push(`CWD: ${meta.cwd}`)
+  if (meta.outputFile) lines.push(`Full output: ${meta.outputFile}`)
+  if (meta.truncated) lines.push(`Output mode: showing stdout/stderr previews; each stream is limited to the last ${COMMAND_PREVIEW_LINES} lines and both streams share ${COMMAND_PREVIEW_TOTAL_CHARS} characters. Full output is saved to the log file.`)
+  if (meta.logError) lines.push(`Log warning: ${meta.logError}`)
+  lines.push('', tailLabel('STDOUT', meta.stdoutTruncated), stdout || '(empty)', '', tailLabel('STDERR', meta.stderrTruncated), stderr || '(empty)')
+  return lines.join('\n')
+}
+
+function safeLogFilePart(value, fallback) {
+  const text = String(value || '').replace(/[^a-zA-Z0-9_.-]/g, '_').slice(0, 80)
+  return text || fallback
+}
+
+async function createCommandLogStream(command, { cwd, description, toolCallId } = {}) {
+  const dir = path.join(logsDir, 'commands')
+  await fs.mkdir(dir, { recursive: true })
+  const timestamp = new Date().toISOString().replace(/[:.]/g, '-')
+  const fileName = `${timestamp}_${safeLogFilePart(toolCallId, 'command')}.log`
+  const outputFile = path.join(dir, fileName)
+  const stream = createWriteStream(outputFile, { flags: 'wx' })
+  stream.write([
+    `Command: ${command}`,
+    description ? `Description: ${description}` : null,
+    `CWD: ${cwd}`,
+    `Started at: ${new Date().toISOString()}`,
+    '',
+  ].filter(Boolean).join('\n'))
+  return { stream, outputFile }
+}
+
+function writeCommandLog(stream, source, chunk) {
+  stream.write(`\n[${source} ${new Date().toISOString()}]\n`)
+  stream.write(chunk)
+}
+
+function killProcessTree(child, signal = 'SIGTERM') {
+  if (!child?.pid) return
+
+  if (process.platform === 'win32') {
+    const killer = spawn('taskkill', ['/pid', String(child.pid), '/T', '/F'], {
+      stdio: 'ignore',
+      windowsHide: true,
+    })
+    killer.on('error', () => {
+      try { child.kill(signal) } catch { /* ignore */ }
+    })
+    return
+  }
+
+  try {
+    process.kill(-child.pid, signal)
+  } catch {
+    try { child.kill(signal) } catch { /* ignore */ }
+  }
+}
+
+const runningCommands = new Map()
+
+export function abortRunningCommand(toolCallId) {
+  if (!toolCallId) return false
+  const stop = runningCommands.get(toolCallId)
+  if (!stop) return false
+  stop('abort')
+  return true
+}
+
+export async function toolRunCommand(params, context, runtime = {}) {
+  const command = String(params?.command || '')
+  if (!command.trim()) {
+    const error = new Error('command is required')
+    error.statusCode = 400
+    throw error
+  }
+
+  const description = String(params?.description || '').trim().slice(0, 500)
+  const timeoutMs = clampNumber(params?.timeoutMs, DEFAULT_RUN_COMMAND_TIMEOUT_MS, MIN_RUN_COMMAND_TIMEOUT_MS, MAX_RUN_COMMAND_TIMEOUT_MS)
+  const cwd = getToolWorkspaceRoot(context)
+  const startedAt = Date.now()
+
+  if (runtime.signal?.aborted) {
+    const details = {
+      command,
+      description,
+      project: context?.project,
+      cwd,
+      timeoutMs,
+      outputFile: null,
+      stdout: '',
+      stderr: 'Command aborted before start.',
+      stdout_preview: '',
+      stderr_preview: 'Command aborted before start.',
+      stdoutTruncated: false,
+      stderrTruncated: false,
+      stdout_truncated: false,
+      stderr_truncated: false,
+      outputTruncated: false,
+      truncated: false,
+      previewLineLimit: COMMAND_PREVIEW_LINES,
+      previewCharLimit: COMMAND_PREVIEW_TOTAL_CHARS,
+      previewMode: 'tail',
+      durationMs: 0,
+      aborted: true,
+    }
+    const content = formatCommandOutput(command, details.stdout, details.stderr, details)
+    return { content: truncateText(content), details }
+  }
+
+  let logStream = null
+  let outputFile = null
+  let logError = null
+  try {
+    const log = await createCommandLogStream(command, { cwd, description, toolCallId: runtime.toolCallId })
+    logStream = log.stream
+    outputFile = log.outputFile
+    logStream.on('error', (error) => {
+      logError = error?.message || 'Failed to write command log.'
+    })
+  } catch (error) {
+    logError = error?.message || 'Failed to create command log.'
+  }
+
+  return new Promise((resolve) => {
+    const child = spawn(command, {
+      cwd,
+      shell: true,
+      stdio: ['ignore', 'pipe', 'pipe'],
+      windowsHide: true,
+      detached: process.platform !== 'win32',
+    })
+
+    let stdout = ''
+    let stderr = ''
+    let stdoutTruncated = false
+    let stderrTruncated = false
+    let timedOut = false
+    let aborted = false
+    let settled = false
+    let updateTimer = null
+    let updatePending = false
+    let forceKillTimer = null
+
+    const cleanup = () => {
+      clearTimeout(timer)
+      if (forceKillTimer) clearTimeout(forceKillTimer)
+      if (updateTimer) clearTimeout(updateTimer)
+      if (runtime.toolCallId) runningCommands.delete(runtime.toolCallId)
+      runtime.signal?.removeEventListener?.('abort', onAbort)
+    }
+
+    const commonDetails = (extra = {}) => {
+      const now = Date.now()
+      const previews = balanceCommandPreviews(stdout, stderr)
+      const stdoutPreview = previews.stdout
+      const stderrPreview = previews.stderr
+      const stdoutPreviewTruncated = stdoutTruncated || previews.stdoutTruncated
+      const stderrPreviewTruncated = stderrTruncated || previews.stderrTruncated
+      const truncated = stdoutPreviewTruncated || stderrPreviewTruncated
+      return {
+        command,
+        description,
+        project: context?.project,
+        cwd,
+        timeoutMs,
+        outputFile,
+        stdout: stdoutPreview,
+        stderr: stderrPreview,
+        stdout_preview: stdoutPreview,
+        stderr_preview: stderrPreview,
+        stdoutTruncated: stdoutPreviewTruncated,
+        stderrTruncated: stderrPreviewTruncated,
+        stdout_truncated: stdoutPreviewTruncated,
+        stderr_truncated: stderrPreviewTruncated,
+        outputTruncated: truncated,
+        truncated,
+        previewLineLimit: COMMAND_PREVIEW_LINES,
+        previewCharLimit: COMMAND_PREVIEW_TOTAL_CHARS,
+        previewMode: 'tail',
+        durationMs: now - startedAt,
+        toolCallId: runtime.toolCallId,
+        logError,
+        ...extra,
+      }
+    }
+
+    const resolveAfterLogClose = (result) => {
+      if (!logStream) {
+        resolve(result)
+        return
+      }
+      const details = result.details || {}
+      logStream.write([
+        '',
+        '',
+        `[quickforge ${new Date().toISOString()}]`,
+        `Exit code: ${details.code ?? 'unknown'}${details.signal ? `, signal: ${details.signal}` : ''}`,
+        `Duration: ${formatDurationMs(details.durationMs)}`,
+        `Timed out: ${Boolean(details.timedOut)}`,
+        `Aborted: ${Boolean(details.aborted)}`,
+        'Command finished.',
+        '',
+      ].join('\n'))
+      logStream.end(() => resolve(result))
+    }
+
+    const finish = ({ code = null, signal = null, error = null } = {}) => {
+      if (settled) return
+      flushUpdate()
+      settled = true
+      cleanup()
+      const durationMs = Date.now() - startedAt
+      if (error) {
+        const details = commonDetails({ error: error.message, aborted, timedOut, durationMs })
+        resolveAfterLogClose({
+          isError: true,
+          content: truncateText(formatCommandOutput(command, details.stdout, `Error running command: ${error.message}\n${details.stderr}`.trim(), details)),
+          details,
+        })
+        return
+      }
+      const details = commonDetails({ code, signal, timedOut, aborted, durationMs })
+      const content = formatCommandOutput(command, details.stdout, details.stderr, details)
+      resolveAfterLogClose({ content: truncateText(content), details })
+    }
+
+    const stopChild = (reason) => {
+      if (reason === 'timeout') timedOut = true
+      if (reason === 'abort') aborted = true
+      killProcessTree(child, 'SIGTERM')
+      forceKillTimer = setTimeout(() => {
+        killProcessTree(child, 'SIGKILL')
+      }, 1500)
+    }
+
+    if (runtime.toolCallId) runningCommands.set(runtime.toolCallId, stopChild)
+
+    function onAbort() {
+      stopChild('abort')
+      finish({ signal: 'SIGTERM' })
+    }
+
+    const runningDetails = () => commonDetails({ running: true })
+
+    const emitUpdate = () => {
+      updateTimer = null
+      if (settled || !updatePending) return
+      updatePending = false
+      const details = runningDetails()
+      runtime.onUpdate?.({
+        content: [{ type: 'text', text: truncateText(formatCommandOutput(command, details.stdout, details.stderr, details)) }],
+        details,
+      })
+    }
+    const flushUpdate = () => {
+      if (updateTimer) {
+        clearTimeout(updateTimer)
+        updateTimer = null
+      }
+      if (!updatePending) return
+      updatePending = false
+      const details = runningDetails()
+      runtime.onUpdate?.({
+        content: [{ type: 'text', text: truncateText(formatCommandOutput(command, details.stdout, details.stderr, details)) }],
+        details,
+      })
+    }
+    const scheduleUpdate = () => {
+      if (settled) return
+      updatePending = true
+      if (!updateTimer) updateTimer = setTimeout(emitUpdate, 150)
+    }
+    const timer = setTimeout(() => {
+      stopChild('timeout')
+      finish({ signal: 'SIGTERM' })
+    }, timeoutMs)
+
+    runtime.signal?.addEventListener?.('abort', onAbort, { once: true })
+
+    child.stdout.on('data', (chunk) => {
+      if (settled) return
+      const text = chunk.toString()
+      const nextStdout = tailText(stdout, text)
+      stdoutTruncated = stdoutTruncated || nextStdout.truncated
+      stdout = nextStdout.text
+      if (logStream && !logError) writeCommandLog(logStream, 'stdout', text)
+      scheduleUpdate()
+    })
+    child.stderr.on('data', (chunk) => {
+      if (settled) return
+      const text = chunk.toString()
+      const nextStderr = tailText(stderr, text)
+      stderrTruncated = stderrTruncated || nextStderr.truncated
+      stderr = nextStderr.text
+      if (logStream && !logError) writeCommandLog(logStream, 'stderr', text)
+      scheduleUpdate()
+    })
+    child.on('close', (code, signal) => {
+      finish({ code, signal })
+    })
+    child.on('error', (err) => {
+      finish({ error: err })
+    })
+  })
+}
+
+export const toolHandlers = {
+  read_file: toolReadFile,
+  grep_files: toolGrepFiles,
+  write_file: toolWriteFile,
+  edit_file: toolEditFile,
+  run_command: toolRunCommand,
+  activate_skill: toolActivateSkill,
+  read_skill_resource: toolReadSkillResource,
+}