context/role.md
role.md is the foundation file. Without it, drafts default to a generic professional voice and triage decisions ignore the differences between work that sits in your remit and work that doesn’t.
It’s also the easiest of the context files to write. Most of the content already exists in your head — you just need to put it on the page.
What to include
Section titled “What to include”- Your title and team
- Who you report to, and the cadence of that relationship
- Your scope — products you own, products you influence, regions, client segments
- What you’re measured on (outcomes, not activities)
- What’s explicitly outside your remit (saves Claude drafting “yes I can help” replies to things you should redirect)
- Your working pattern — time zone, core hours, days you’re typically heads-down
What to exclude
Section titled “What to exclude”- Company-confidential strategy details
- Salary, performance ratings, anything sensitive about your employment
- Anything you’d be uncomfortable surfacing if a draft accidentally got pasted somewhere unintended
- Aspirational scope (“I want to influence X”) — Claude needs your real scope, not your goal
Template
Section titled “Template”# Role
## Position[Title] in [team] at [company].
## Reports to[Name], [Title]. We have a 1:1 every [day/cadence].
## Scope- Products/areas I own: [list]- Products/areas I influence but don't own: [list]- Regions: [list]- Client segments I work with: [issuers / merchants / both, tier 1 / tier 2, etc]
## What I'm measured on- [Outcome 1, e.g. revenue from advisory engagements]- [Outcome 2, e.g. number of clients using a specific product]- [Outcome 3, e.g. thought leadership reach]
## What's NOT my remit[Things people sometimes ask me about that should go elsewhere.This section saves Claude drafting "yes I can help with that"replies to things I should redirect.]
## Working pattern- Time zone: [zone]- Core hours: [hours]- Days I'm typically heads-down: [days, if any]Why “what’s NOT my remit” matters
Section titled “Why “what’s NOT my remit” matters”The most common failure mode of AI-drafted replies is over-promising. Without explicit scope boundaries, Claude defaults to helpful-and-accommodating, and produces drafts that commit you to work you shouldn’t be doing.
Listing what’s outside your remit gives Claude a clear signal: when an email asks for something on this list, the right draft is a redirect, not a yes. Two or three lines here saves dozens of edits over time.
How often to update
Section titled “How often to update”Every time your role changes — new title, new manager, new scope, new measurement. In practice this means once or twice a year, plus an annual review pass to catch drift.