SkillsandAgents.co just published Equity reconciliation via Carta, a skill that reconciles your Carta cap table against your HRIS employee list and flags gaps: mismatched termination dates, terminated employees still holding active grants, lapsed exercise windows, and grants missing source documents. It reads live data directly from the source system without requiring CSV exports.
Skills as instructions
A skill is a set of instructions. It tells the model what to pull, how to match employees across two systems, how to rank exceptions by severity, and what to write in the report. It does not contain or store a cap table, employee list, or any names. The information in the skill is not stored in memory.
MCP-native means the data stays in your account
MCP is the Model Context Protocol, the standard way for an application to expose its data to an AI client. Carta runs an MCP server. When you use this skill, your client connects to that server using your OAuth login, scoped to read only the Cap Table plugin. The model pulls your stakeholder list, grants, and vesting data live, at runtime, straight from your Carta account into your session, so reconciliation happens without exports.
The CSV export
The platform-agnostic version of this skill, equity-reconciliation, works from a CSV export of your cap table. That is fine and the right tool for teams not on Carta. But since a CSV export is a static file full of names and equity data, every copy risks PII leakage.
Going MCP-native removes that file. There is no export to misplace, no stale copy on a shared drive, no attachment in a thread. Because the skill uses MCP, the workflow can run as often as needed with a slash command, keeping the process live and export-free.
When to run it
Run it on demand when you want a clean picture, on a schedule (quarterly is common), and always before a capital raise, 409A valuation, or audit. It requires the Carta MCP server connected to your client and an HRIS export with at least employee ID, email, name, status, and termination date. The setup and full field list are on the skill page.
If you want a skill like this built into your own systems and privacy constraints, that is the work we do.
For the operator’s view on why we design skills to read live data and hold none of yours, see edakrong.com.