book-sft-pipeline
This skill should be used when the user asks to "fine-tune on books", "create SFT dataset", "train style model", "extract ePub text", or mentions style transfer, LoRA training, book segmentation, or author voice replication.
72
61%
Does it follow best practices?
Impact
92%
1.95xAverage score across 3 eval scenarios
Advisory
Suggest reviewing before use
Optimize this skill with Tessl
npx tessl skill review --optimize ./examples/book-sft-pipeline/SKILL.mdQuality
Discovery
72%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
The description excels at providing trigger terms and occupying a distinct niche, but critically lacks an explicit statement of what the skill actually does. It reads more like a list of when-to-use conditions than a proper skill description. Adding a clear capability statement before the trigger clause would significantly improve it.
Suggestions
Add an explicit 'what it does' statement at the beginning, e.g., 'Extracts text from ePub files, segments books into passages, and generates SFT-format training datasets for LoRA fine-tuning and author style replication.'
Restructure to lead with capabilities in third person ('Processes ePub books to create...') followed by the existing 'Use when...' trigger clause.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | The description mentions several actions like 'fine-tune on books', 'create SFT dataset', 'extract ePub text', and concepts like 'book segmentation' and 'author voice replication', but these are embedded within trigger phrases rather than stated as concrete capabilities the skill performs. There's no clear statement of what the skill actually does (e.g., 'Extracts text from ePub files and segments it into training pairs for SFT/LoRA fine-tuning'). | 2 / 3 |
Completeness | The description has a strong 'when' clause with explicit trigger phrases, but the 'what does this do' part is essentially missing — there is no clear statement of the skill's capabilities or outputs. The 'what' can only be inferred from the trigger terms. Per the rubric, this is 'Has what, but when is missing or only implied' inverted — it has when but what is only implied. | 2 / 3 |
Trigger Term Quality | The description includes a strong set of natural trigger terms that users would actually say: 'fine-tune on books', 'create SFT dataset', 'train style model', 'extract ePub text', 'style transfer', 'LoRA training', 'book segmentation', 'author voice replication'. These cover both technical and natural language variations well. | 3 / 3 |
Distinctiveness Conflict Risk | The combination of ePub extraction, SFT dataset creation, LoRA training, book segmentation, and author voice replication is a very specific niche. It's highly unlikely to conflict with other skills given the specialized domain of book-based fine-tuning and style transfer. | 3 / 3 |
Total | 10 / 12 Passed |
Implementation
50%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The skill excels at actionability with complete, executable code for every pipeline phase and concrete examples throughout. However, it is significantly bloated by the 'Integration with Context Engineering Skills' section (~60 lines of conceptual mapping with zero actionable content), redundant explanations of concepts Claude already understands, and metadata that doesn't serve the skill's purpose. The workflow would benefit from intermediate validation checkpoints between phases rather than only end-stage validation.
Suggestions
Remove or drastically reduce the 'Integration with Context Engineering Skills' section — it provides no actionable guidance and wastes ~60 lines of context window on conceptual mappings.
Add intermediate validation checkpoints between phases (e.g., verify extraction produced clean text before segmenting, verify chunk word counts meet 150-400 target, verify instruction quality before dataset construction).
Remove the 'Core Concepts' section — the three pillars are already demonstrated in the phase implementations and the guidelines section; explaining them separately is redundant.
Move 'Known Issues and Solutions', cost estimates, and expected results to a separate reference file to keep the main skill lean and focused on the executable pipeline.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is extremely verbose at ~400+ lines. The 'Integration with Context Engineering Skills' section (~60 lines) adds no actionable value and merely maps concepts to other skills. The 'Core Concepts' section explains things Claude already knows (e.g., why breaking mid-sentence is bad). The pipeline architecture ASCII diagram, while nice, is redundant with the detailed phase descriptions. Metadata, cost estimates, and expected results tables add bulk with marginal utility. | 1 / 3 |
Actionability | The skill provides fully executable Python code for every phase: extraction, segmentation, instruction generation, dataset construction, and training. Code examples are complete and copy-paste ready with specific libraries, parameters, and data formats including concrete JSON message structures. | 3 / 3 |
Workflow Clarity | The six phases are clearly sequenced and well-labeled, but there are no explicit validation checkpoints between phases. For example, there's no step to verify extraction quality before segmentation, no check that chunks meet word count targets, and no feedback loop if instruction generation produces poor descriptions. The validation phase exists only at the end, not as intermediate checkpoints. | 2 / 3 |
Progressive Disclosure | References to external files (segmentation-strategies.md, tinker-format.md, tinker.txt) are well-signaled at the end, but the main file itself is monolithic with too much inline content. The 'Integration with Context Engineering Skills' section and the 'Known Issues' section could be separate files. The core content that should remain inline (~phases 1-6) is still quite long but reasonably structured. | 2 / 3 |
Total | 8 / 12 Passed |
Validation
90%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 10 / 11 Passed
Validation for skill structure
| Criteria | Description | Result |
|---|---|---|
frontmatter_unknown_keys | Unknown frontmatter key(s) found; consider removing or moving to metadata | Warning |
Total | 10 / 11 Passed | |
- Commit
7a95d94
- 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.