Prompt files and skills
When a slash-command prompt file is enough and when a skill earns its folder — building /new-engagement and the meeting-prep skill.
- Tell a prompt file apart from a skill and know when each is right
- Build a reusable slash command for scaffolding an engagement
- Package a repeatable task as a skill with its own resources
Two mechanisms turn a good prompt into a reusable one, and the choice between
them is a small but real design decision. Prompt files are .prompt.md files you
invoke as slash commands⊙ . Skills are
SKILL.md folders — an Anthropic-originated open standard that Copilot adopted —
discovered from .github/skills/ and enabled by default across the editor, the
CLI, and the cloud agent.⊙
When a prompt file is enough
If the reusable thing is just instructions — do these steps, in this order, with this output — a prompt file is the right tool. It is one file, it shows up as a slash command, and it carries no baggage. Scaffolding a new engagement is exactly this shape:
---
description: Scaffold a new engagement repo from the template and personalize the instruction files.
mode: agent
---
# The Demo Harness — drewbreyer.com/harness — starter file, as of 2026-07-06
Scaffold a new customer engagement in this workspace.
1. Create the numbered artifact folders: 00-intake, 01-research, 02-discovery,
03-model, 04-data, 05-agents, 06-rehearsal, 07-leave-behind.
2. Copy the .github/ instruction stack, agents, prompts, skills, and hooks, plus
.vscode/mcp.json and AGENTS.md, from the template.
3. Ask me for the customer name, the opportunity, and the compete, then
personalize copilot-instructions.md with those specifics.
4. Leave 00-intake/ with a README that lists what may and may not live in the
repo (public research and my notes yes; customer data of record no).
Report the tree you created and the questions you still need answered.
Invoked as /new-engagement, it creates the numbered folders, copies the
instruction stack, and personalizes the constitution by asking you for the
specifics. No resources, no scripts — just a well-worn procedure, captured.
When a skill earns its folder
A skill earns the extra folder when the task needs to carry resources alongside the instructions — a template to fill, a checklist to apply, a script to run. The meeting-prep task is a skill because it ships a fixed one-page brief template and a pre-meeting checklist, and the judgment about how to assemble a brief lives with those resources:
---
name: meeting-prep
description: Assemble a one-page pre-meeting brief from the dossier, the findings, and pasted calendar context.
---
# The Demo Harness — drewbreyer.com/harness — starter file, as of 2026-07-06
# Meeting prep
Produce a single page a seller can read in the elevator.
## When to use
Before any customer meeting once 01-research/ and 02-discovery/ exist. This is a
skill rather than a prompt file because it carries a fixed brief template and a
short checklist as resources.
## Steps
1. Read 01-research/dossier.md and 02-discovery/findings.md.
2. Take the meeting's attendees and agenda from what the user pastes (calendar
context is not in this repo).
3. Fill brief-template.md: who is in the room and what each cares about, the
three things we believe, the two questions we must ask, and the one risk.
4. Keep it to one page. Cite each product claim; leave anything uncertain out.
## Resources
- brief-template.md — the one-page layout.
- checklist.md — the pre-meeting read.
The test is simple: if deleting everything but the instructions would lose nothing, write a prompt file. If it would lose a template or a script the task depends on, write a skill.
In the field
Take a procedure you run every engagement and write it as a /-command prompt
file. Then find one that carries a template you always reach for, and promote it
to a skill folder with that template beside it. Invoke each once to feel the
difference.
- .github/prompts/new-engagement.prompt.md — the scaffold command
- .github/skills/meeting-prep/SKILL.md — the meeting-prep skill