dsh-tool-encoding

中文

DSH encoding/hash tool plugin — base64/base64url/url/hex encode/decode + hash + UUID for UTF-8 text. Zero dependencies, zero subprocesses, pure functions.

Package name: @deepseek-ai/dsh-tool-encoding (standalone bundle, not a monorepo-integrated form); the lib/ output is generated by npm run build (tsc) inside the repo and committed with the repository.

Motivation

Encoding/hashing is a high-frequency daily operation for agents: inspecting base64 fields in API responses (JWT payload), building query parameters, verifying file integrity (sha256), generating UUIDs. Problems with the current approach bash -c "echo ... | base64": subprocess overhead, quote-escaping hell (a base64 string embedded in a bash command then in a JSON argument — the double-escaping error rate is extremely high), and inconsistent cross-platform tool names (md5 vs md5sum vs shasum -a 256).

Security model

No eval, no new Function. All operations are pure-function compositions of Buffer/node:crypto/TextDecoder/encodeURIComponent:

• UTF-8 integrity: decoding uses fatal mode (TextDecoder('utf-8', { fatal: true })); illegal bytes throw encoding: invalid UTF-8 output; legal U+FFFD/control characters are not rejected (00 NUL, 0a newline, efbfbd U+FFFD are all legal, ff is illegal)
• Strict base64 validation: no whitespace, = only at the end and ≤2, length must be a multiple of 4, RFC 4648 canonical unused bits (non-canonical encodings such as Zh== are rejected, preventing multiple strings from mapping to the same text)
• Unified rejection of lone surrogates: all text inputs pass through String.prototype.isWellFormed(), avoiding silent replacement and inconsistent URIError behavior
• Byte limits: 1 MB input / 4 MB output (1,000,000 / 4,000,000 bytes each, Buffer.byteLength; the output limit is a fuse of pre-allocation estimate + final check)
• Hash algorithm whitelist: Object.hasOwn lookup table (md5/sha1/sha256/sha512)
• Errors uniformly use the encoding: prefix and never pass through the underlying URIError/TypeError

Architecture

DSH Agent
    │ ctx.tools.register()
    ▼
src/index.ts（Cordis 插件入口 + action 分发 + 独立校验）
    │
    ▼
src/encoding.ts
    ├── b64Encode/b64Decode — 标准 base64（严格校验）
    ├── b64UrlEncode/b64UrlDecode — base64url（canonical 无 padding，解码兼容）
    ├── urlEncode/urlDecode — component 语义（URIError 包装）
    ├── hexEncode/hexDecode — UTF-8 字节 hex（fatal 解码）
    ├── digest — 白名单哈希
    ├── newUuid — crypto.randomUUID()
    └── validateUnicode/assertInputBytes/decodeUtf8Strict — 校验器

Tool declaration

ctx.tools.register(defineTool({
  name: 'encoding',
  parameters: {
    action: {
      type: 'string', required: true,
      enum: ['base64_encode','base64_decode','base64url_encode','base64url_decode',
             'url_encode','url_decode','hex_encode','hex_decode','hash','uuid'],
    },
    input:     { type: 'string', description: 'Input string for encode/decode/hash' },
    algorithm: { type: 'string', enum: ['md5','sha1','sha256','sha512'], description: 'For hash' },
  },
  output: { schema: { type: 'json' }, render: (_a, v) => [{ type: 'text', text: JSON.stringify(v) }] },
  execute: (args) => Promise.resolve(executeAction(args.action, args) as JsonValue),
  timeoutMs: 1000,
}))

Supported operations

Semantic contract:

• v1 is a UTF-8 text tool: all inputs and outputs are strings; binary content is represented as hex; decoded results must be valid UTF-8
• All actions return strings (including uuid)
• Do not use this tool on confidential material: tool arguments are recorded in session logs

DSH 0.1.5-rc.1 compatibility (verified)

This plugin has been migrated to the DSH 0.1.5-rc.1 harness and fully verified in an isolated consumer of local harness 0.1.5-rc.1:

• Types/runtime: @deepseek-ai/cordis@^4.0.1 + @deepseek-ai/dsh-tools@>=0.0.1-rc.1 <0.2.0 + @deepseek-ai/dsh-invariants@>=0.0.1-rc.1 <0.2.0 (peer); no longer depends on unscoped cordis
• Standalone build: npm install (devDependencies are self-contained: typescript/vitest/@types/node) → npm run typecheck → npm test → npm run build → npm pack
• Consumption verification: tarball installed into the 0.1.5-rc.1 consumer → dsh --profile compat --dump-config shows this plugin's row → the tool actually registers and executes
• Startup: npx -p @deepseek-ai/dsh@next dsh web (lib production mode; do not install -g globally)

Version adaptation

• DSH version adapted: DSH 0.1.5-rc.1
• Bundle declaration: dsh.bundle in package.json (patch points to cordis.patch.yml) + exports fields
• Patch format: cordis.patch.yml uses the - insert: list (patches are id-targeted; a bare - id: entry reports entry not found)
• files: the published tarball contains lib/, src/, cordis.patch.yml

Installation

Plugin source repository: https://github.com/omdsh-dev/dsh-tool-encoding (public).

Profile Bundle (recommended)

Install this plugin as a standalone bundle into a profile (DSH 0.1.5-rc.1, npm):

# 交互式（web）profile
dsh plugin --profile web add github:omdsh-dev/dsh-tool-encoding
# 一次性任务（headless）profile —— dsh run 默认使用 headless
dsh plugin --profile headless add github:omdsh-dev/dsh-tool-encoding

The dsh.bundle.patch inside the package (pointing to cordis.patch.yml) automatically adds the plugin to the profile's layer stack after installation; the plugin's cordis.patch.yml inserts the tool-encoding entry via - insert:.

⚠️ web and headless are different profiles: installing into web does not automatically cover headless; dsh run uses the headless profile by default.

Install via npm pack tarball

npm pack    # generates dsh-tool-encoding-*.tgz
dsh plugin --profile web add ./dsh-tool-encoding-*.tgz
dsh plugin --profile headless add ./dsh-tool-encoding-*.tgz

Verify installation

dsh --profile web --dump-config | grep tool-encoding

Run verification

dsh run "使用 encoding 工具把 hello 做 base64 编码"

Manual installation and legacy compatibility

Only for legacy snapshots that do not support Profile Bundle, or plugin development/debugging environments:

1. Place into the monorepo: cp -r encoding ~/.dsh/source/master/packages/tools/encoding (development/debugging)
2. Add "@deepseek-ai/dsh-tool-encoding": "workspace:^" to apps/cli/package.json; add { "path": "./packages/tools/encoding" } to the references of tsconfig.host.json
3. pnpm install && pnpm run build
4. Insert the plugin in the profile's user-layer patch (~/.dsh/profiles/<name>/cordis.patch.yml):

- insert:
    - id: tool-encoding
      name: '@deepseek-ai/dsh-tool-encoding'

1. Verify: dsh --profile <name> --dump-config | grep tool-encoding

Note: patches are id-targeted — a bare - id: entry reports entry "xxx" not found; it must be wrapped in an - insert: list.

Known limitations

1. Arbitrary binary (non-printable bytes) encode/decode requires v2's output: "utf8" | "hex" mode
2. hash provides digests only, no HMAC/salting/key derivation; MD5/SHA-1 are for compatibility/non-security integrity checks only
3. URL uses component semantics; form encoding (space → +) needs a separate action (v2)

Testing

pnpm test

41 test cases covering functionality/errors/attack payloads (RFC 4648/1321/6234 known vectors, UTF-8 boundary distinctions, canonical padding boundaries, unified surrogate rejection, etc.). See the locally maintained design document for the full list.

License

MIT