CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

translator

Offline translation across ~419 languages via MADLAD-400-3B-MT (CTranslate2 + SentencePiece). No cloud. No account. The model downloads once (~3 GB) and runs locally on CPU or Apple Silicon thereafter. Use whenever the user wants to translate text TO or FROM a non-English language — short phrases, paragraphs, whole documents, idioms, low-resource languages, indigenous languages, dead languages MADLAD has seen, polite forms, casual registers. Also trigger on "translate", "say this in <lang>", "what does <foreign text> mean", "render this in <language>", "<lang> for <phrase>", "how do you say X in Y", "localize", or any request that involves moving text between human languages. DO NOT use for code translation between programming languages. DO NOT use for transliteration (script conversion) without a target meaning — MADLAD translates meaning, not glyphs.

Quality

82%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

61%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured, largely actionable body whose Workflow, Tool reference, and Calibration sections carry genuinely non-obvious operational detail backed by real bundled scripts. Its weaknesses are a batch mode with no output-verification guidance despite documented hallucination risk, editorial padding (Sovereignty notes, a stray trailing line, description duplication), and a couple of references to files not present in the bundle.

Suggestions

Add a verification step for --batch output — e.g., spot-check a few translated lines (especially for low-resource targets where hallucination is documented) before handing results back — so the batch workflow has a feedback loop.

Trim the Sovereignty notes section to 2-3 lines, remove the stray trailing 'enough-tooltip-text:' line, and cut the 'When to use' overlap with the frontmatter description to recover token budget.

Fix the dangling references: bundle the requirements.txt that 'pip install -r requirements.txt' depends on, and document how translate.py selects the 7B variant after 'bootstrap.py --install-7b' (no such flag or mechanism is shown).

DimensionReasoningScore

Conciseness

The operational core (Workflow, Tool reference, Calibration notes on quality tiers, load-time amortization, and paragraph splitting) is dense and carries model-specific knowledge Claude does not have. But there is real padding that could be trimmed: the ~19-line 'Sovereignty notes' section ('on a laptop on a plane, in a tent, in a SCIF', 'deliberate implementation of enough's "no cloud dependency, no SDK lock-in" principle') is editorial rationale rather than execution guidance, the Integration section states the never-through-the-LLM rule twice, 'When to use' substantially duplicates the frontmatter description, and a stray dangling 'enough-tooltip-text:' line closes the body. This fits anchor 3 ('some unnecessary explanation or could be tightened') better than 4's 'minor instances', since the trimmable content spans several sections.

3 / 5

Actionability

Commands are copy-paste ready and all documented flags (--check, --install, --install-7b, --status, --where, --uninstall, --target, --batch, --list-langs) verified as real in the bundled scripts; the three ordered invocation patterns cover the common cases well. Minor gaps keep it from 5: 'pip install -r requirements.txt' references a requirements.txt that is not present in the bundle, and the 7B variant can be downloaded (--install-7b) but no documented way exists to make translate.py actually use it.

4 / 5

Workflow Clarity

The core sequence is clear and includes an explicit pre-flight checkpoint with exit-code branching ('bootstrap.py --check' → 'If the exit code is non-zero, install it'), which alone would suggest 4-5. However, the skill documents a first-class batch mode ('--batch < strings.txt' for UI string lists) and itself warns that low-resource pairs 'may produce stilted output, repeat tokens, or hallucinate', yet gives no verification step for batch output (e.g., spot-checking lines before use). Per the rubric cap for batch operations without validation, workflow clarity cannot exceed 3.

3 / 5

Progressive Disclosure

Good structure: sections are well-organized, both bundle scripts exist and get an inline CLI reference at the right altitude, and external pointers ('rness/paradigms/translation.md', 'rness/knowledge/rosetta-primers/') are one level deep and clearly signaled with licensing context. Not 5 because two referenced paths (requirements.txt, the rness/ tree) are not verifiable within the bundle — a dangling reference and externally-dependent navigation — and the calibration material, while useful, is all inlined with no clear split for the bulk detail.

4 / 5

Total

14

/

20

Passed

Description

95%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: third-person, concrete, honest about scope, with an explicit use-when clause, natural trigger phrasings, and explicit anti-triggers that sharply reduce conflict risk. The only soft spot is that the capability list is one core action expanded across input types rather than several distinct operations.

DimensionReasoningScore

Specificity

Concrete and specific — 'Offline translation across ~419 languages via MADLAD-400-3B-MT (CTranslate2 + SentencePiece)' names the exact model and stack, and the scope covers 'short phrases, paragraphs, whole documents, idioms, low-resource languages, indigenous languages, dead languages MADLAD has seen, polite forms, casual registers'. It falls short of the 5 anchor because the action list is essentially one verb (translate) applied to many input types, unlike the anchor-5 example's multiple distinct operations; it exceeds 3 because coverage plus explicit exclusions goes well beyond 1-2 generic actions.

4 / 5

Completeness

Explicitly answers both: what — 'Offline translation across ~419 languages via MADLAD-400-3B-MT' with the download/local-execution model spelled out; when — 'Use whenever the user wants to translate text TO or FROM a non-English language' with concrete trigger phrases. It also adds explicit negative triggers ('DO NOT use for code translation between programming languages', 'DO NOT use for transliteration'), which is above the anchor-5 example.

5 / 5

Trigger Term Quality

Comprehensive natural trigger phrasing: 'translate', 'say this in <lang>', 'what does <foreign text> mean', 'render this in <language>', '<lang> for <phrase>', 'how do you say X in Y', 'localize', plus the catch-all 'any request that involves moving text between human languages'. These are phrases a user would naturally say; no meaningful synonyms are missing.

5 / 5

Distinctiveness Conflict Risk

Clear niche — offline natural-language translation via a named model — with explicit boundary markers distinguishing it from code translation between programming languages and from transliteration/script conversion. Minimal risk of triggering for the wrong skill; the 'dead languages MADLAD has seen' phrasing honestly scopes the claim.

5 / 5

Total

19

/

20

Passed

Validation

100%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository

0gsd/enough

Path

defaults/skills/translator/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.