A skill is a packaged, repeatable procedure that an AI agent follows to carry out a recurring work process. Each one defines its inputs, steps, outputs, and known failure modes. It also records the reason behind each rule, so the procedure improves as problems are found instead of being rediscovered.

Research & knowledge management

Research into Knowledge Base

Conducts parallel research on a topic and files the results as linked notes, with entities and concepts cross-referenced in both directions.

1. Topic question
2. Parallel research
3. Synthesized report
4. Linked notes

Meetings & communication

Meeting Notes

Converts recorded meeting summaries and transcripts into structured notes with action items. It can be re-run safely without creating duplicate notes or tasks.

1. Recorded meeting
2. Summary & transcript
3. Structured note
4. Action items

Meeting Prep

Builds a meeting agenda from a template, adding relevant context drawn from earlier notes.

1. Upcoming meeting
2. Agenda template
3. Prior context
4. Ready agenda

Morning Brief

Produces a daily briefing covering the day's meetings, tasks due, and urgent items.

1. Calendar
2. Open tasks
3. Prioritize
4. Daily brief

Presentation & visual communication

Document to Voice

Converts written documents into narrated audio, so the same content can be listened to as well as read.

1. Written document
2. Prepared script
3. Speech synthesis
4. Narrated audio

Canvas Math

Renders real mathematical notation on a visual canvas, from either structured math markup or LaTeX.

1. Math markup
2. Rendered formula
3. Canvas element

Interactive Review Pages

Turns a set of static pages into a review surface. Reviewers leave comments in place, and the comments flow back to the agent for revision.

1. Static pages
2. Reviewer comments
3. Comment queue
4. Agent revision

Evaluation & quality review

Local Model Evaluation

Compares AI models head to head on domain-specific test suites (see Evaluation).

1. Test suite
2. One model at a time
3. Scored results
4. Comparison

Local Model Operations

Covers running, configuring, and choosing among local models by task, including a catalogue of known failure modes.

1. Task type
2. Model selection
3. Local run
4. Failure-mode check

Document Evaluator

Runs a six-stage pipeline that assesses large document collections and recommends a treatment for each document.

1. Collection
2. Profile & assess
3. Classify
4. Recommendation

Interface Scanner

Runs automated, element-level quality scans of web pages and reports each issue with its location.

1. Web page
2. Automated scan
3. Element issues
4. Report

Software Review

Researches the usability and standards conformance of software under consideration for purchase, and produces a conformance report.

1. Product
2. Research
3. Assessment
4. Conformance report