dsh-ssh-workbench 0.0.0-stage → 1.7.0

package/lib/core/remote-file-tools.js

@@ -0,0 +1,559 @@
+/**
+ * Explicit-target remote file tools.
+ *
+ * Shell-only remote access makes ordinary troubleshooting possible, but a
+ * single `cat`/`sed`/`grep` command cannot express "read this file", "replace
+ * this unique string" or "list files matching a glob" without brittle shell
+ * quoting. These tools add a small, deliberately non-transparent file layer:
+ * every call names its managed resource, reads and writes go through SFTP,
+ * writes are gated by configuration, and every operation is capped.
+ *
+ * They are additive conveniences, not a remote IDE: the native local
+ * `read/write/edit/glob/grep` tools keep their local semantics.
+ */
+import { defineTool } from '@deepseek-ai/dsh-tools'
+
+import { describeResultOutcome } from './activity.js'
+import { resolveDialect, sessionKeyOf } from './remote-shell-tool.js'
+import { requireServerRef } from './server-ref.js'
+import { win32Adapter } from './win32-adapter.js'
+
+export const REMOTE_READ_TOOL_NAME = 'remote_read'
+export const REMOTE_WRITE_TOOL_NAME = 'remote_write'
+export const REMOTE_EDIT_TOOL_NAME = 'remote_edit'
+export const REMOTE_GLOB_TOOL_NAME = 'remote_glob'
+export const REMOTE_GREP_TOOL_NAME = 'remote_grep'
+
+const textOut = {
+  schema: { type: 'object', additionalProperties: false, properties: { text: { type: 'string', required: true } } },
+  render: (_args, value) => [{ type: 'text', text: value.text }],
+}
+
+const MAX_READ_HARD_LIMIT = 64 * 1024 * 1024
+const MAX_EDIT_HARD_LIMIT = 8 * 1024 * 1024
+const MAX_RESULTS_HARD_LIMIT = 2000
+const MAX_DEPTH_HARD_LIMIT = 12
+
+function describeError(error) {
+  return error instanceof Error ? error.message : String(error)
+}
+
+function integerIn(value, fallback, min, max) {
+  return Number.isInteger(value) ? Math.max(min, Math.min(value, max)) : fallback
+}
+
+function loadServer({ loadInventory, serverRef, toolName }) {
+  let servers
+  try {
+    servers = loadInventory()
+  } catch (error) {
+    throw new Error(`cannot read the managed server inventory: ${describeError(error)}`)
+  }
+  return requireServerRef(servers, serverRef, { label: `${toolName} server` })
+}
+
+function normalizeRemotePath(value, dialect) {
+  let path = String(value ?? '').trim()
+  if (!path) throw new Error('path is required')
+  if (path.includes('\0')) throw new Error('path must not contain NUL bytes')
+  if (path === '~' || path.startsWith('~/') || path.startsWith('~\\')) {
+    return { home: true, path: path.slice(1) }
+  }
+  if (!(path.startsWith('/') || path.startsWith('\\\\') || /^[A-Za-z]:[\\/]/.test(path))) {
+    throw new Error(`path must be absolute or start with ~: ${path}`)
+  }
+  if (dialect === 'windows') path = path.replace(/\\/g, '/')
+  return { home: false, path }
+}
+
+async function resolveRemotePath({ runtime, server, rawPath, dialect }) {
+  const normalized = normalizeRemotePath(rawPath, dialect)
+  let resolved
+  if (normalized.home) {
+    let home
+    try {
+      home = await runtime.sftpHome(server)
+    } catch (error) {
+      throw new Error(`cannot resolve the remote home directory: ${describeError(error)}`)
+    }
+    resolved = normalized.path ? `${home.replace(/[\\/]+$/, '')}/${normalized.path.replace(/^[\\/]+/, '')}` : home
+  } else {
+    resolved = normalized.path
+  }
+  return dialect === 'windows' ? resolved.replace(/\\/g, '/') : resolved
+}
+
+function joinRemotePath(parent, child) {
+  return `${String(parent ?? '').replace(/[\\/]+$/, '')}/${String(child ?? '').replace(/^[\\/]+/, '')}`
+}
+
+function joinRelative(parent, child) {
+  return parent ? `${parent}/${child}` : child
+}
+
+function globToRegExp(pattern) {
+  let source = '^'
+  const value = String(pattern ?? '').replace(/\\/g, '/').replace(/^\.\//, '')
+  for (let index = 0; index < value.length; index += 1) {
+    const char = value[index]
+    if (char === '*') {
+      if (value[index + 1] === '*') {
+        index += 1
+        if (value[index + 1] === '/') {
+          index += 1
+          source += '(?:.*/)?'
+        } else {
+          source += '.*'
+        }
+      } else {
+        source += '[^/]*'
+      }
+    } else if (char === '?') {
+      source += '[^/]'
+    } else {
+      source += /[.+^${}()|[\]\\]/.test(char) ? `\\${char}` : char
+    }
+  }
+  source += '$'
+  return new RegExp(source)
+}
+
+function matchesGlob(pattern, relativePath) {
+  const normalized = String(pattern ?? '').replace(/\\/g, '/').replace(/^\.\//, '')
+  const target = normalized.includes('/') ? relativePath : relativePath.split('/').at(-1)
+  return globToRegExp(normalized).test(target)
+}
+
+function renderHeader({ server, path, extra = [] }) {
+  const target = `${server.username ?? 'root'}@${server.host}:${server.port ?? 22}`
+  return [`target: ${server.name} (${target})`, path === undefined ? null : `path: ${path}`, ...extra].filter(
+    (line) => line !== null,
+  )
+}
+
+function describeActivityOutcome(outcome) {
+  return typeof outcome === 'string' && outcome.length > 0 ? outcome : 'unknown'
+}
+
+function recordFileActivity({ recordActivity, server, sessionId, startedAt, command, outcome }) {
+  if (typeof recordActivity !== 'function') return
+  try {
+    recordActivity({
+      kind: 'exec',
+      server: server.name,
+      command,
+      outcome,
+      sessionId,
+      durationMs: Date.now() - startedAt,
+    })
+  } catch {
+    // Activity is best effort and must never fail a file operation.
+  }
+}
+
+function assertWritesEnabled(config) {
+  if (config?.remoteFileWrite !== true) {
+    throw new Error(
+      'remote file writes are disabled by configuration (remoteFileWrite=false); enable them in the plugin config before using remote_write or remote_edit',
+    )
+  }
+}
+
+function byteLength(value) {
+  return Buffer.byteLength(String(value ?? ''), 'utf8')
+}
+
+export function createRemoteReadTool({ runtime, config, loadInventory, recordActivity }) {
+  return defineTool({
+    name: REMOTE_READ_TOOL_NAME,
+    description:
+      'Read a text file from a managed remote resource through SFTP. The path must be absolute or start with ~. Binary files and files over the read limit are rejected/truncated; use the workbench SFTP panel for large or binary downloads.',
+    parameters: {
+      server: { type: 'string', required: true, description: 'Managed resource name or id.' },
+      path: { type: 'string', required: true, description: 'Absolute remote path, or a path under ~.' },
+      maxBytes: { type: 'number', description: 'Optional read cap in bytes (clamped by plugin config).' },
+    },
+    output: textOut,
+    isConcurrencySafe: () => true,
+    async execute(args, exec) {
+      const startedAt = Date.now()
+      const server = loadServer({ loadInventory, serverRef: args?.server, toolName: 'remote_read' })
+      const sessionId = sessionKeyOf(exec)
+      try {
+        const dialect = await resolveDialect(server, runtime)
+        const path = await resolveRemotePath({ runtime, server, rawPath: args?.path, dialect })
+        const cap = integerIn(config?.remoteFileReadBytes, 262144, 1024, MAX_READ_HARD_LIMIT)
+        const maxBytes = integerIn(args?.maxBytes, cap, 1024, cap)
+        const result = await runtime.sftpRead(server, path, { maxBytes })
+        if (result?.binary === true) throw new Error(`remote file is binary: ${path}`)
+        const lines = renderHeader({
+          server,
+          path,
+          extra: [
+            `size: ${result?.size ?? 'unknown'} bytes`,
+            `read: ${result?.bytesRead ?? 0} bytes`,
+            result?.truncated === true ? '[truncated by the configured read limit]' : null,
+          ].filter(Boolean),
+        })
+        lines.push('content:')
+        const content = typeof result?.content === 'string' ? result.content : ''
+        lines.push(content.length === 0 ? '(empty)' : content.replace(/\n$/, ''))
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_read ${path}`,
+          outcome: 'success',
+        })
+        return { text: lines.join('\n') }
+      } catch (error) {
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_read ${String(args?.path ?? '')}`,
+          outcome: `failed (${describeError(error)})`,
+        })
+        throw error
+      }
+    },
+  })
+}
+
+export function createRemoteWriteTool({ runtime, config, loadInventory, recordActivity }) {
+  return defineTool({
+    name: REMOTE_WRITE_TOOL_NAME,
+    description:
+      'Write a UTF-8 text file on a managed remote resource through SFTP. Requires remoteFileWrite to be enabled in the plugin config. The parent directory must already exist; the file is created or overwritten.',
+    parameters: {
+      server: { type: 'string', required: true, description: 'Managed resource name or id.' },
+      path: { type: 'string', required: true, description: 'Absolute remote path, or a path under ~.' },
+      content: { type: 'string', required: true, description: 'Complete UTF-8 file content.' },
+    },
+    output: textOut,
+    isConcurrencySafe: () => false,
+    async execute(args, exec) {
+      assertWritesEnabled(config)
+      const startedAt = Date.now()
+      const server = loadServer({ loadInventory, serverRef: args?.server, toolName: 'remote_write' })
+      const sessionId = sessionKeyOf(exec)
+      try {
+        const dialect = await resolveDialect(server, runtime)
+        const path = await resolveRemotePath({ runtime, server, rawPath: args?.path, dialect })
+        const content = String(args?.content ?? '')
+        const bytes = byteLength(content)
+        const cap = integerIn(config?.remoteFileEditBytes, 2097152, 1024, MAX_EDIT_HARD_LIMIT)
+        if (bytes > cap) throw new Error(`file content is ${bytes} bytes, over the ${cap}-byte write limit`)
+        const result = await runtime.sftpWrite(server, path, content)
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_write ${path}`,
+          outcome: 'success',
+        })
+        return {
+          text: renderHeader({
+            server,
+            path,
+            extra: [`wrote: ${result?.bytes ?? bytes} bytes`],
+          }).join('\n'),
+        }
+      } catch (error) {
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_write ${String(args?.path ?? '')}`,
+          outcome: `failed (${describeError(error)})`,
+        })
+        throw error
+      }
+    },
+  })
+}
+
+export function createRemoteEditTool({ runtime, config, loadInventory, recordActivity }) {
+  return defineTool({
+    name: REMOTE_EDIT_TOOL_NAME,
+    description:
+      'Replace text in a UTF-8 file on a managed remote resource. By default oldText must match exactly once; set replaceAll=true to replace every occurrence. Requires remoteFileWrite to be enabled.',
+    parameters: {
+      server: { type: 'string', required: true, description: 'Managed resource name or id.' },
+      path: { type: 'string', required: true, description: 'Absolute remote path, or a path under ~.' },
+      oldText: { type: 'string', required: true, description: 'Exact text to find.' },
+      newText: { type: 'string', required: true, description: 'Replacement text.' },
+      replaceAll: { type: 'boolean', description: 'Replace every occurrence instead of requiring a unique match.' },
+    },
+    output: textOut,
+    isConcurrencySafe: () => false,
+    async execute(args, exec) {
+      assertWritesEnabled(config)
+      const startedAt = Date.now()
+      const server = loadServer({ loadInventory, serverRef: args?.server, toolName: 'remote_edit' })
+      const sessionId = sessionKeyOf(exec)
+      try {
+        const dialect = await resolveDialect(server, runtime)
+        const path = await resolveRemotePath({ runtime, server, rawPath: args?.path, dialect })
+        const oldText = String(args?.oldText ?? '')
+        if (!oldText) throw new Error('oldText must not be empty')
+        const newText = String(args?.newText ?? '')
+        const cap = integerIn(config?.remoteFileEditBytes, 2097152, 1024, MAX_EDIT_HARD_LIMIT)
+        const file = await runtime.sftpRead(server, path, { maxBytes: cap })
+        if (file?.binary === true) throw new Error(`remote file is binary: ${path}`)
+        if (file?.truncated === true) throw new Error(`remote file is larger than the ${cap}-byte edit limit: ${path}`)
+        const source = typeof file?.content === 'string' ? file.content : ''
+        const occurrences = source.split(oldText).length - 1
+        if (occurrences === 0) throw new Error(`oldText was not found in ${path}`)
+        if (occurrences > 1 && args?.replaceAll !== true) {
+          throw new Error(`oldText occurs ${occurrences} times in ${path}; pass a longer unique string or set replaceAll=true`)
+        }
+        const next = args?.replaceAll === true ? source.split(oldText).join(newText) : source.replace(oldText, newText)
+        if (byteLength(next) > cap) throw new Error(`edited file would exceed the ${cap}-byte write limit`)
+        await runtime.sftpWrite(server, path, next)
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_edit ${path}`,
+          outcome: 'success',
+        })
+        return {
+          text: renderHeader({
+            server,
+            path,
+            extra: [`replacements: ${args?.replaceAll === true ? occurrences : 1}`],
+          }).join('\n'),
+        }
+      } catch (error) {
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_edit ${String(args?.path ?? '')}`,
+          outcome: `failed (${describeError(error)})`,
+        })
+        throw error
+      }
+    },
+  })
+}
+
+async function walkRemote({ runtime, server, root, pattern, maxDepth, maxResults }) {
+  const results = []
+  const queue = [{ directory: root, relative: '', depth: 0 }]
+  let visited = 0
+  while (queue.length > 0 && results.length < maxResults && visited < maxResults * 20) {
+    const current = queue.shift()
+    let entries
+    try {
+      entries = await runtime.sftpList(server, current.directory)
+    } catch (error) {
+      if (current.depth === 0) throw error
+      continue
+    }
+    for (const entry of Array.isArray(entries) ? entries : []) {
+      if (!entry || entry.type === 'link' || typeof entry.name !== 'string') continue
+      visited += 1
+      const relative = joinRelative(current.relative, entry.name)
+      const path = entry.path ?? joinRemotePath(current.directory, entry.name)
+      if (entry.type === 'dir') {
+        if (current.depth < maxDepth) queue.push({ directory: path, relative, depth: current.depth + 1 })
+      } else if (matchesGlob(pattern, relative)) {
+        results.push({ path, relative, type: entry.type })
+        if (results.length >= maxResults) break
+      }
+    }
+  }
+  return { results, truncated: results.length >= maxResults }
+}
+
+export function createRemoteGlobTool({ runtime, config, loadInventory, recordActivity }) {
+  return defineTool({
+    name: REMOTE_GLOB_TOOL_NAME,
+    description:
+      'Find files on a managed remote resource by glob through SFTP. Supports *, ? and **. Search path defaults to the remote home; symlinks are not followed; depth and result count are capped.',
+    parameters: {
+      server: { type: 'string', required: true, description: 'Managed resource name or id.' },
+      pattern: { type: 'string', required: true, description: 'Glob pattern, e.g. "**/*.log" or "etc/*.conf".' },
+      path: { type: 'string', description: 'Directory to search; defaults to the remote home directory.' },
+      maxResults: { type: 'number', description: 'Optional result cap (clamped by plugin config).' },
+      maxDepth: { type: 'number', description: 'Optional recursion depth (clamped by plugin config).' },
+    },
+    output: textOut,
+    isConcurrencySafe: () => true,
+    async execute(args, exec) {
+      const startedAt = Date.now()
+      const server = loadServer({ loadInventory, serverRef: args?.server, toolName: 'remote_glob' })
+      const sessionId = sessionKeyOf(exec)
+      try {
+        const dialect = await resolveDialect(server, runtime)
+        const root = args?.path
+          ? await resolveRemotePath({ runtime, server, rawPath: args.path, dialect })
+          : await resolveRemotePath({ runtime, server, rawPath: '~', dialect })
+        const pattern = String(args?.pattern ?? '').trim()
+        if (!pattern) throw new Error('pattern is required')
+        const maxResults = integerIn(args?.maxResults, config?.remoteGlobMaxResults ?? 200, 1, MAX_RESULTS_HARD_LIMIT)
+        const maxDepth = integerIn(args?.maxDepth, config?.remoteGlobMaxDepth ?? 4, 0, MAX_DEPTH_HARD_LIMIT)
+        const walked = await walkRemote({ runtime, server, root, pattern, maxDepth, maxResults })
+        const lines = renderHeader({
+          server,
+          path: root,
+          extra: [`pattern: ${pattern}`, `matches: ${walked.results.length}`],
+        })
+        if (walked.results.length === 0) lines.push('(no matches)')
+        else for (const match of walked.results) lines.push(match.path)
+        if (walked.truncated) lines.push('[results truncated by the configured limit]')
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_glob ${pattern}`,
+          outcome: 'success',
+        })
+        return { text: lines.join('\n') }
+      } catch (error) {
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_glob ${String(args?.pattern ?? '')}`,
+          outcome: `failed (${describeError(error)})`,
+        })
+        throw error
+      }
+    },
+  })
+}
+
+function shellQuote(value) {
+  return `'${String(value ?? '').replace(/'/g, "'\\''")}'`
+}
+
+function powershellQuote(value) {
+  return `'${String(value ?? '').replace(/'/g, "''")}'`
+}
+
+function posixGrepCommand({ path, pattern, ignoreCase, maxResults }) {
+  const flags = ['-R', '-I', '-n', '-E']
+  if (ignoreCase === true) flags.push('-i')
+  return [
+    `grep ${flags.join(' ')} --exclude-dir=.git --exclude-dir=node_modules -e ${shellQuote(pattern)} -- ${shellQuote(path)}`,
+    `head -n ${maxResults}`,
+  ].join(' | ')
+}
+
+function windowsGrepScript({ path, pattern, ignoreCase, maxResults }) {
+  const caseSwitch = ignoreCase === true ? '' : ' -CaseSensitive'
+  return [
+    `$path = ${powershellQuote(path)}`,
+    `$item = Get-Item -LiteralPath $path -ErrorAction SilentlyContinue`,
+    `if ($null -eq $item) { Write-Error ('path not found: ' + $path); exit 3 }`,
+    `if ($item.PSIsContainer) { $files = @(Get-ChildItem -LiteralPath $path -Recurse -File -ErrorAction SilentlyContinue) } else { $files = @($item) }`,
+    `$files | Select-String -Pattern ${powershellQuote(pattern)}${caseSwitch} | Select-Object -First ${maxResults} | ForEach-Object { "$($_.Path):$($_.LineNumber):$($_.Line)" }`,
+  ].join('\n')
+}
+
+export function createRemoteGrepTool({ runtime, config, loadInventory, recordActivity }) {
+  return defineTool({
+    name: REMOTE_GREP_TOOL_NAME,
+    description:
+      'Search text in files on a managed remote resource. Uses grep on POSIX targets and Select-String on Windows targets; the path may be a file or directory and results are capped. Binary files are skipped on POSIX.',
+    parameters: {
+      server: { type: 'string', required: true, description: 'Managed resource name or id.' },
+      pattern: { type: 'string', required: true, description: 'Pattern to search for (POSIX extended regex / PowerShell regex).' },
+      path: { type: 'string', description: 'File or directory to search; defaults to the remote home directory.' },
+      ignoreCase: { type: 'boolean', description: 'Case-insensitive search.' },
+      maxResults: { type: 'number', description: 'Optional result cap (clamped by plugin config).' },
+    },
+    output: textOut,
+    isConcurrencySafe: () => true,
+    async execute(args, exec) {
+      const startedAt = Date.now()
+      const server = loadServer({ loadInventory, serverRef: args?.server, toolName: 'remote_grep' })
+      const sessionId = sessionKeyOf(exec)
+      try {
+        const dialect = await resolveDialect(server, runtime)
+        const path = args?.path
+          ? await resolveRemotePath({ runtime, server, rawPath: args.path, dialect })
+          : await resolveRemotePath({ runtime, server, rawPath: '~', dialect })
+        const pattern = String(args?.pattern ?? '')
+        if (!pattern) throw new Error('pattern is required')
+        if (pattern.includes('\n') || pattern.includes('\r')) throw new Error('pattern must not contain newlines')
+        const maxResults = integerIn(args?.maxResults, config?.remoteGrepMaxResults ?? 200, 1, MAX_RESULTS_HARD_LIMIT)
+        let command
+        if (dialect === 'windows') {
+          const built = win32Adapter.buildCommand(
+            windowsGrepScript({ path, pattern, ignoreCase: args?.ignoreCase === true, maxResults }),
+            undefined,
+            server,
+          )
+          if (built?.error !== undefined) throw new Error(built.error)
+          command = built.command
+        } else {
+          // Validate the root first so a missing path is an error rather than
+          // an empty "no matches" result.
+          await runtime.sftpStat(server, path)
+          command = posixGrepCommand({ path, pattern, ignoreCase: args?.ignoreCase === true, maxResults })
+        }
+        const result = await runtime.exec(server, command, {
+          timeoutMs: config?.commandTimeoutMs ?? 120000,
+          signal: exec?.signal,
+          readOnly: true,
+          cap: config?.maxOutputBytes ?? 102400,
+          stderrCap: config?.maxStderrBytes ?? 16384,
+        })
+        const exitCode = Number.isInteger(result?.exitCode) ? result.exitCode : null
+        if (exitCode !== null && exitCode > 1 && dialect !== 'windows') {
+          throw new Error(`remote grep failed (exit ${exitCode}): ${result?.stderr || 'no stderr'}`)
+        }
+        const stdout = typeof result?.stdout === 'string' ? result.stdout : ''
+        const lines = renderHeader({
+          server,
+          path,
+          extra: [`pattern: ${pattern}`, `matches: ${stdout.trim() ? stdout.trim().split('\n').length : 0}`],
+        })
+        lines.push(stdout.trim() ? stdout.replace(/\n$/, '') : '(no matches)')
+        if (result?.truncated?.stdout === true) lines.push('[stdout truncated by the plugin output cap]')
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_grep ${pattern}`,
+          outcome: describeResultOutcome(result),
+        })
+        return { text: lines.join('\n') }
+      } catch (error) {
+        recordFileActivity({
+          recordActivity,
+          server,
+          sessionId,
+          startedAt,
+          command: `remote_grep ${String(args?.pattern ?? '')}`,
+          outcome: `failed (${describeError(error)})`,
+        })
+        throw error
+      }
+    },
+  })
+}
+
+/** Build all five definitions in the canonical registration order. */
+export function createRemoteFileTools({ runtime, config, loadInventory, recordActivity }) {
+  return [
+    createRemoteReadTool({ runtime, config, loadInventory, recordActivity }),
+    createRemoteWriteTool({ runtime, config, loadInventory, recordActivity }),
+    createRemoteEditTool({ runtime, config, loadInventory, recordActivity }),
+    createRemoteGlobTool({ runtime, config, loadInventory, recordActivity }),
+    createRemoteGrepTool({ runtime, config, loadInventory, recordActivity }),
+  ]
+}