AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Translation Workbench

skill-alexu0317-father-translation-workbench-translation-workbench · by Alexu0317-FATHER

Initialize or continue a source-grounded translation project through source preparation, terminology alignment, drafting, and user-led finalization. Use when the user names translation-workbench or asks for a structured translation workflow. Do not use for one-off translations, publishing, or post-publication formatting.

— No reviews yet
0 installs
0 views
— view→install

Install

$ agentstack add skill-alexu0317-father-translation-workbench-translation-workbench

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • ✓ Prompt-injection patterns
  • ✓ Secret / credential exfiltration
  • ✓ Dangerous shell & filesystem operations
  • ✓ Untrusted network calls
  • ✓ Known-malicious package signatures

What it can access

  • ✓ Network access No
  • ✓ Filesystem access No
  • ✓ Shell / process execution No
  • ✓ Environment & secrets No
  • ✓ Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-alexu0317-father-translation-workbench-translation-workbench)

Reliability & compatibility

✓ Security review passed
0 installs to date
— no reviews yet
● today

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Translation Workbench? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Translation Workbench

Guide a translation project from source material to a user-approved final translation. Adapt to the user's languages, material, document structure, and current stage instead of assuming a particular genre, book layout, chapter numbering system, or output channel.

User interaction

  • Follow the user's request. Do not force a setup questionnaire, a fixed opening message, or a particular next stage.
  • Inspect the current request and available project files before asking questions. Ask only for missing information that materially changes the work.
  • When a decision has a small, clear set of options, prefer a structured user-input tool exposed by the current runtime. Examples include Claude Code's AskUserQuestion, Codex's request_user_input, or an equivalent tool.
  • Use ordinary conversation for paths, document uploads, long explanations, terminology evidence, translation choices, and other open-ended answers.
  • If no structured user-input tool is available, ask a concise plain-text question.
  • Never treat a tool-selection response as authorization for a different file mutation, external action, or publication.

Start from the user's situation

  1. Read a project-level README.md when one exists and appears to describe the translation project.
  2. Determine whether the user is creating a new project, adopting existing material, starting a new translation unit, or continuing a named stage.
  3. Accept any user-specified work, section, chapter, scene, or other translation unit. Never assume the user starts with the first unit.
  4. Preserve the user's existing directory structure when it is workable. For a new project, use the recommended structure in [project initialization](references/project-initialization.md).
  5. Treat user-supplied source files and reference documents as read-only inputs. Create derived working documents from the templates without overwriting the originals.
  6. Create glossary, character, background, source, or style documents only when real content exists for them. Do not create empty project files merely because templates are available.
  7. After initialization, continue only as far as the user's request requires.

For initialization, adoption of existing material, or recovery from an incomplete setup, read [project-initialization.md](references/project-initialization.md).

Route the requested stage

  • Explain or plan: Explain the relevant part of the workflow without creating files unless the user asks.
  • Prepare source material: Read [sourcing.md](references/sourcing.md).
  • Draft a translation: Read [translation.md](references/translation.md).
  • Merge decisions and finalize: Read [finalization.md](references/finalization.md).

When the user explicitly starts a cross-unit review of completed translations to improve project references, use the separate translation-distillation skill. Do not start it as part of a unit's finalization.

If a requested stage lacks its required inputs, report the missing input and stop that stage. Do not silently reconstruct an earlier stage from guesses.

Session boundaries

  • Recommend a separate session for each major stage because it limits context carryover.
  • Do not require, enforce, or claim to validate how the user organizes sessions.
  • Do not promise that a complete project will behave consistently when run in one long session or across untested models.
  • At the end of a completed stage, state what was produced and give a short suggested prompt for starting the next stage in a new session.
  • A new session must rely on the project README and saved stage artifacts, not on access to the previous conversation.

Deterministic checks

Use the bundled read-only checkers instead of relying on a model assertion when their inputs are available:

  • scripts/check_translation_context.py validates a unit's JSON handoff, checks durable terminology against the glossary, verifies selected Markdown sections, and returns the exact context to read.
  • scripts/check_stage.py checks stage prerequisites, protects an existing finalization record, and confirms that a new finalization begins from the saved initial draft.

Resolve `` from this skill's installed location. Resolve project files from the explicit project root or paths recorded in the project README.

python /scripts/check_translation_context.py --project-root  --handoff 
python /scripts/check_stage.py  --project-root  [stage-specific paths]

Treat status: ready with an empty warnings array as a passed context gate. material_incomplete, terms_pending, or blocked means stop the requested stage and report what remains. error means the input or checker invocation must be corrected.

Invariants

  • Keep source meaning, order, scope, and uncertainty intact. Do not add background information to the translation merely because it appears in reference material.
  • Distinguish source text and verified facts, user-confirmed choices, and AI-written work notes. An AI note records what was considered; it does not establish that the choice was correct.
  • Search existing project terminology before proposing a new durable term.
  • Add or change durable glossary, character, background, or translator-style guidance only after the user has approved the underlying judgment.
  • Keep chapter- or passage-specific choices in drafting or review notes instead of turning them into universal project rules.
  • During finalization, apply non-mechanical changes only after recording the user's decision.
  • Never overwrite an existing finalization record, initial-draft snapshot, or user-supplied source file without explicit authorization.
  • Preserve the initial draft before finalization changes it. Record where the snapshot can be retrieved for a later, user-started distillation.
  • Do not generate publication packages, platform-specific formatting, dashboards, or analytics.
  • Do not claim that behavior or translation quality has been validated for an untested model or language pair.

Templates

Templates are under assets/templates/. Adapt their language and headings to the user's working language while preserving their information roles.

  • project-readme.md: project entry point and file map
  • glossary.md: durable, user-approved terminology
  • character-profiles.md: character and voice guidance
  • background-notes.md: background facts and provenance
  • translator-style.md: durable translation style decisions
  • sources.md: source and research inventory
  • sourcing-handoff.json: machine-readable source-preparation handoff
  • drafting-notes.md: passage-level translation choices
  • review-notes.md: paragraph-by-paragraph finalization decisions and reasons

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.