Skill Builder Guide
A practical guide to designing reusable, triggerable, maintainable Agent Skills.
The material is primarily instructional and shows no malware, credential theft, or covert exfiltration. Its scripts create directories, write files, and package archives, but they do not specify least privilege, user confirmation, sensitive-data handling, rollback, or data-flow disclosure, so points are deducted. Apache-2.0 licensing and official organizational provenance are present in repository context, but the skill itself references LICENSE.txt, leaving attribution and license application incomplete.
The structure and workflow are broadly coherent, covering initialization, editing, packaging, and iteration. However, the PyYAML dependency is undeclared, scripts have no committed tests, key paths and abnormal-input handling lack diagnostic feedback, and static review cannot establish execution, so the score is capped below 10 and reduced.
The trigger description clearly targets creating or updating Claude skills, and the body covers resource types, trigger descriptions, and representative workflows. It does not sufficiently define non-fit cases, environment requirements, Chinese-language support, or semantic false-trigger boundaries, so points are deducted.
The information architecture is readable, uses progressive disclosure, and provides linked references, examples, and resource-organization guidance. It lacks versioning, changelog, maintenance ownership, update path, installation dependencies, and systematic troubleshooting; the LICENSE.txt reference is also uncertain, so points are deducted.
The skill is directly usable for the core task of designing and packaging skills, with process guidance, templates, and examples. It provides no representative verified outputs, user feedback, or execution evidence, so outcome completeness and marginal benefit over manual work remain uncertain; under static calibration the score does not exceed 7.
The files provide auditable workflows, scripts, and repository CI configuration, but the CI primarily validates plugins rather than this skill's key behaviors. No committed test suite, third-party reproduction, or corroborating evidence types are shown; static review therefore supports at most 5 points.
- This is a low-confidence static review; scripts were not executed and packaging results were not independently verified.
- Confirm dependencies such as PyYAML and add tests covering initialization, validation, and packaging paths before relying on it.
- The scripts write files and create directories; confirm target paths, permissions, and overwrite behavior in restricted or production environments.
- Clarify the actual location of LICENSE.txt, maintenance ownership, versioning and update procedures, and support for Chinese users and target network environments.
What it does & when to use it
Skill Builder Guide is the skill-creator component inside the financial-analysis plugin. It helps users create or update skills that extend Claude with specialized knowledge, workflows, and tool integrations. The guide covers SKILL.md design, reusable scripts, references, assets, progressive disclosure, packaging, and iteration. It is intended for developers building custom skills, rather than users seeking a ready-made financial analysis workflow.
Explains how to define a skill through concrete usage examples and trigger scenarios; plan reusable scripts, references, and assets; initialize a new skill with scripts/init_skill.py; write YAML frontmatter containing name and description plus the skill body; validate and package a distributable .skill file with scripts/package_skill.py; and improve the skill through real-world iteration.
- A developer needs to create a new Agent Skill for a specialized domain and wants to clarify realistic usage patterns first.
- A team wants to turn repeatedly rewritten scripts, API documentation, or templates into reusable skill resources.
- A maintainer needs guidance on trigger descriptions, directory structure, and progressive loading of references.
- A skill author has finished editing and needs to validate and package the skill for distribution.
- An existing skill needs refinement after users encounter workflow or efficiency problems.
Pros & cons
- Covers the full workflow from understanding requirements through packaging and iteration.
- Provides clear guidance on concise instructions, trigger descriptions, and progressive disclosure.
- Explains when scripts, reference files, and assets are useful.
- It is a methodology guide, not a ready-made financial analysis capability.
- The supplied material does not show a dedicated test suite or standalone execution examples for this skill.
- Initialization and packaging require the repository's Python scripts and filesystem access.
How to install
The skill is located at plugins/vertical-plugins/financial-analysis/skills/skill-creator/SKILL.md. The README documents Claude Code installation by running claude plugin marketplace add anthropics/financial-services, followed by claude plugin install financial-analysis@claude-for-financial-services. The supplied source does not document a separate installation command for this individual skill.
How to use
In an environment where the skill is loaded, ask for a new or revised skill, such as “Create a PDF-processing skill with reusable scripts, references, and templates.” The documented creation flow uses scripts/init_skill.py <skill-name> --path <output-directory> for initialization and scripts/package_skill.py <path/to/skill-folder> for packaging. These commands must run in a repository environment containing the scripts. Exact automatic triggering depends on the host client.