How I Designed a Claude Skill for Meeting Action Items
Discover how to build a sub-100-line Claude skill using Markdown and YAML frontmatter to turn rough meeting notes into structured action items and recap emails.
Stock photo for illustration only, not from the actual event
- Built a custom Claude skill in under 100 lines using Markdown and YAML frontmatter.
- Optimized the skill description to include inputs, outputs, and explicit trigger conditions.
- Implemented safety rules like appending (proposed) to inferred dates and assignees.
- Enforced a rigid output structure featuring a clear action item table and follow-up email.
Following a client meeting, the administrative routine is always the same: rereading notes, identifying actual decisions, noting assignees and deadlines, and sending a recap before memories fade. Because this task is repetitive and has clear parameters, it serves as a great use case for a Claude skill. The developer built one consisting of under 100 lines of Markdown, released under an MIT license, and made freely available.
A skill is structured as a folder containing a SKILL.md file. It begins with YAML frontmatter defining a name and description, followed by Markdown instructions. There is no custom code, external dependencies, or API calls involved. Claude simply reads the instructions when the skill matches the relevant context and follows them accordingly.
Stock photo for illustration only, not from the actual event
A critical design choice involves writing a description that helps Claude accurately determine when to trigger the skill. The description explicitly names three elements: inputs (transcripts, rough notes), outputs (summaries, action tables, follow-up emails), and explicit Use when clauses to ensure the phrasing aligns with how users actually make requests.
Designing lightweight modular skills like this highlights the power of structured prompt engineering without heavy backend infrastructure. It allows developers to reliably constrain LLM outputs and tailor AI behavior to specific administrative workflows with minimal overhead.
To prevent hallucinated meeting minutes, the skill enforces strict operational rules:
- Extract only what was spoken without inventing decisions, owners, or dates.
- Mark inferences transparently by appending (proposed) to guessed assignees or deadlines.
- Preserve vague commitments by assigning them an owner of TBD instead of discarding them.
- Keep finalized decisions and open questions strictly separated.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment