Skills

A skill is a set of instructions the agent draws on when a task calls for it — a procedure written down once, rather than pasted into a prompt every time.

What ships

Six workflows come with the product and work in any directory:

Skill	What it does
spreadsheet-review	Checks a file of rows against its stated rules, and reports the rows that break them with their row numbers
weekly-report	What actually changed here over a period — done, in progress, and what needs a decision
inbox-triage	Sorts messages by what they are asking for, and flags the ones a person must answer
document-compare	What changed in substance between two drafts, separately from what only changed in wording
meeting-notes	A transcript into decisions, actions with owners, and open questions
records-review	Raises what is worth checking in a set of records, citing the note each came from

Two more, inherited, are about writing and installing skills themselves: skill-creator and skill-installer.

Where they come from

Scope	Directory
Shipped	~/.opencli/skills/.system — rewritten when the app updates
Yours	~/.opencli/skills
This project's	.opencli/skills in the working directory

Writing one

A directory with a SKILL.md, whose front matter says what it is and when to use it:

---
name: invoice-check
description: Check invoices against our payment rules and report what breaks
  them. Use when asked to review, check or audit a file of invoices.
metadata:
  short-description: Find invoices that break the rules
---

# Invoice check

## Find the rules before reading the rows
...

The description is what the agent matches a request against. A description that only says what the skill is leaves the model to guess when it applies — every shipped skill names its occasions, and yours should too.

What makes one work

Reading the six that ship is the fastest way to see the shape, but three things recur:

• Say what to read, and in what order. "Oldest first" is a real instruction; "review the records" is not.
• Say what to produce. A named file, or a stated structure. Not "a summary".
• Say what it must never do. The shipped clinical skill spends a third of its length on this, and that is why it can exist.

One found the hard way: if a number matters, say where the number must come from. The spreadsheet skill asks for a row count, and the first run stated it from memory and got it wrong by one. It now requires that the count come from the script that did the checking.

Turning one off

Each skill has a switch in Abilities → Skills. A change applies to the next chat you open, not the one already running.