PerryLink / dsh-translate

---

Compatibility

Surface	Status
Harness	DeepSeek Harness 0.1.0-rc.6 (compat declared for 0.1.0-rc.5–0.1.0-rc.6)
Node	^22.19.0 || >=24.0.0
Form	Pure-host JS plugin (no browser half)
Model	Any model — repair is deterministic, no extra model calls

What you get

Two independent surfaces, one bundle:

• /translate — the vendor parameter translation table: temperature, top_p, max_tokens, stop, system, and 8 more canonical parameters mapped across 11 vendors (OpenAI, ERNIE, Qwen, Anthropic, Google, DeepSeek, Mistral, Cohere, xAI, Groq, Azure). Ask for a pairwise mapping, list vendors/params, or convert a whole standard request (transformRequest in lib/rosetta.mjs).
• The repair layer — a tools/post-execute listener plus the fix_json tool. When a successful tool result carries broken JSON as a string (string-rooted or json-rooted output schema, or a tool opted in by name), the layer repairs it deterministically: markdown-fence extraction, escape repair, trailing-comma removal, truncation closure, and required-field completion with explicit null placeholders. No value is ever invented — a result that still violates the schema fails closed, and failed tool results are never flipped into successes.

tool result (success, JSON text) ──▶ extract fence ──▶ parse
    │ ok? ──▶ schema-validate ──▶ accept { kind: 'accept', value }  (registry re-validates + re-renders)
    │ broken ──▶ escape / trailing-comma / close / fill-null ──▶ validate
    │ unrepairable ──▶ next()  (original value preserved) + translate/fix audit (counts only)

Quick start

# 1. install the bundle into your profile
dsh plugin --profile web add "github:PerryLink/dsh-translate#main"

# or from npm (published releases)
dsh plugin --profile web add dsh-translate

# 2. restart and verify the row
dsh --profile web --dump-config | grep -A2 'id: dsh-translate'

Then ask the agent to check a mapping or repair a payload:

> /translate openai ernie max_tokens
> Use fix_json to repair: {"a": 1,} against {"type":"object","properties":{"a":{"type":"integer"}},"required":["a"]}

Install & uninstall

• git channel (latest main): dsh plugin --profile web add "github:PerryLink/dsh-translate#main" — pure JS, no build step.
• npm channel (published releases): dsh plugin --profile web add dsh-translate.
• tarball channel: pnpm pack in this repo, then dsh plugin --profile web add ./dsh-translate-<version>.tgz.
• uninstall: dsh plugin --profile web remove dsh-translate (or remove the row from the profile patch).

Configuration

All tunables are Schemastery Config fields (changeable from cordis.yml). An id-targeted override replaces the whole row — restate every key you need. cordis.patch.yml documents each key inline.

Key	Default	Meaning
enabled	true	Master switch; false registers nothing
repair.enabled	true	Post-execute repair layer switch
repair.toolNames	[]	Extra tool names whose JSON-text results may be repaired (on top of string-rooted / json-rooted schemas)
repair.strategies.escapeRepair	true	Escape raw control characters inside strings
repair.strategies.trailingComma	true	Remove commas directly before a closing bracket
repair.strategies.truncationClosure	true	Close an unclosed string or container cut off by truncation
repair.strategies.fieldCompletion	true	Complete missing required fields with explicit null placeholders
repair.maxSteps	8	Strategy application budget (repair-loop passes, 1..64)
diffMaxChars	200	Cap on one logged diff fragment, in characters
diffMaxEntries	50	Cap on logged diff entries
registerCommand	true	Register the /translate command
registerTool	true	Register the fix_json tool

Example override in your profile patch:

- insert:
    - id: dsh-translate
      name: dsh-translate
      config:
        enabled: true
        repair:
          enabled: true
          toolNames: ['emit-json']
          strategies:
            escapeRepair: true
            trailingComma: true
            truncationClosure: true
            fieldCompletion: true
          maxSteps: 8
        diffMaxChars: 200
        diffMaxEntries: 50
        registerCommand: true
        registerTool: true

Tools & surfaces

Surface	Kind	Notes
/translate	command	vendors, params, or <from> <to> [param] pairwise mapping
fix_json	tool	{ text, schema?, strategies? } → { ok, repaired?, diff?, strategies, truncated, validated, error? }; diff fragments are bounded and sanitized
post-execute repair	listener	Automatic for successful string results from string-rooted / json-rooted schemas (plus repair.toolNames); always calls next() unless it claims the call

Permissions & data

• Permissions: no network, no subprocess, no credentials — the plugin only consumes the official commands and tools services and appends to the session log.
• Data: repair never fabricates values; the only model-visible additions are the repaired canonical value and the fix_json diff. Session audit events (translate/fix) carry tool name, call id, strategy names, edit counts, and truncation flags — never payloads.

Security boundaries

• Deterministic only. Repair is bounded text surgery; the upstream LLM-retry arm of JSON-Schema-Enforcer-Proxy was deliberately not ported — a post-execute listener never calls a model.
• Fail closed. Unrepairable syntax and schema violations leave the original result untouched (or return a structured error from fix_json); null placeholders only land when the schema accepts null.
• Hostile input bounded. Unsupported schema keywords and circular schemas are rejected; oneOf validation is capped by depth (MAX_ONE_OF_DEPTH) and a branch budget (MAX_ONE_OF_BUDGET), so an exponential schema cannot exhaust the process.
• No payload leakage. Logs and audit events never contain repaired payloads; diffs are truncated and capped before display or storage.

Known limitations

• The supported JSON Schema subset mirrors the harness tool registry (type/oneOf/properties/required/additionalProperties/items/enum/const); other keywords are rejected as unsupported, not silently ignored.
• Repair only applies to successful results whose canonical value is a JSON-text string; a value that already failed schema validation arrives as a failed result and is never flipped.
• The translation table covers 11 vendors × 13 canonical parameters; extended rows follow public API references (not the upstream trio) and are marked as such in lib/rosetta.mjs.

Development

pnpm install        # node ^22.19 || >=24
pnpm test           # node --test: 57 tests (pure lib suites + real-service assembly suite)
pnpm run check      # tsc checkJs against types.d.ts
pnpm run verify:self-contained  # dependency specs resolve from the registry
pnpm run verify:artifacts       # ESM face imports under plain Node + lib exports present
node scripts/check-readme-sync.mjs  # five-language README sync gate (also in CI)
pnpm pack           # the published tarball

Topics

dsh, dsh-plugin, deepseek-harness, deepseek, cordis, json-repair, schema-validation, parameter-mapping, llm-api, tooling

Contributors

• @PerryLink — creator and maintainer: translation table and repair pipeline ports, plugin surfaces, tests, and the five-language docs.

PerryLink DSH Plugin Family

This project is one of the 29 DeepSeek Harness plugins maintained by PerryLink. If this one helps you, the others likely will too:

Plugin	One-liner
dsh-auto-review	Second-model auto-review on the approval chain, fail-closed by default
dsh-background-agents	Durable background child agents with a Web UI sidebar, messaging and interrupt
dsh-budget	Cost governance for DeepSeek Harness: budgets, carbon, and latency in one panel.
dsh-checkpoint-rewind	Claude Code /rewind-equivalent: snapshots, session forks, one-shot restore
dsh-claude-move	Migrate Claude Code sessions, memory, skills and CLAUDE.md into DSH
dsh-click	Cross-platform native desktop control for DeepSeek Harness — Windows first.
dsh-composer-history	Terminal-style input history for the web composer: arrows, Ctrl+R search
dsh-defend	Prompt-injection, jailbreak, and secret-leak defense for DeepSeek Harness.
dsh-doublecheck	Engineering-discipline guard: requirements grill, test gates, adversary review
dsh-draw	Unified static-image generation routing for DeepSeek Harness.
dsh-fast	Read-only performance diagnostics for DeepSeek Harness.
dsh-github	GitHub PR/issues integration for DSH, every write gated by approval
dsh-library	Local document knowledge base for DeepSeek Harness.
dsh-local-ai	Local-model (Ollama) integration for DeepSeek Harness.
dsh-lsp-actions	LSP diagnostics, formatting, completion, code actions and rename over language servers
dsh-mask	PII masking middleware for DeepSeek Harness — anonymize personal data before it reaches the model, restore it at the display layer.
dsh-mcp-panel	Read-only MCP runtime panel: /mcp command + Settings tab with status, tools and errors
dsh-memento	Approval-gated cross-session memory: ctx.memory seam + SQLite + memory tool
dsh-observe	OpenTelemetry and Langfuse observability exporter for DeepSeek Harness.
dsh-output-styles	Claude Code outputStyles-equivalent runtime style switching
dsh-permission-rules	Claude Code-style declarative allow/deny/ask permission rules with audit
dsh-plugin-guide	Plugin-development knowledge base as an on-demand agent skill
dsh-score	Multi-dimensional quality scoring for DeepSeek Harness plugins.
dsh-session-pin	Pin sessions in the Web sidebar with durable ordering
dsh-session-sync	Cross-device session sync for DeepSeek Harness — a dedicated git mirror of your session store.
dsh-skill-pack-security	Security-audit skill pack: secret scan, dependency and supply-chain review
dsh-talk	Voice-first session loop for DeepSeek Harness: talk to it, hear it answer.
dsh-test-drive	Isolated install-and-smoke test drives for DeepSeek Harness plugins.
dsh-translate	Vendor parameter translation and deterministic JSON repair for DeepSeek Harness.

License

Apache License 2.0 © 2026 dsh-translate contributors