Every routine below works with a plain prompt and the folder above. Once a prompt earns its place, save it as a slash command (see the last section).
Every morning: the brief
What: a five-minute read of what changed and what you need to decide today.
Read Notes/ from yesterday and my open tickets in Linear. List: what
changed, anything blocked on me, and the three decisions I should make
today, each with the file or ticket it came from. Save to
Notes/<today>-brief.md.
Pulling tickets needs an MCP connection to Linear or Jira (MCP guide). Without one, export or paste the list.
Pitfall: a brief that lists everything. Cap it at three decisions or you'll stop reading it by Wednesday.
Before meetings: prep and a pre-read
What: who's in the room, what each person wants, the objections you'll hear, and the one outcome you need.
I'm meeting the eng lead and head of sales about moving saved searches
to next quarter. Using Context/ and Work/decisions.md, write a one-page
prep: what each person optimizes for, their likely objections, my ask,
and a fallback. Save to Work/meetings/.
In AI PM OS this routine ships as /meeting and /stakeholder.
Pitfall: treating the predicted objections as fact. It's a rehearsal, not a read on people.
After meetings: capture decisions
What: turn messy notes into decisions, owners and risks, and update context so next week's drafts know.
Here are my notes from the roadmap review. Extract decisions (what,
who decided, date), action items with owners, and new risks. Append to
Work/decisions.md. If a decision changes goals.md or constraints.md,
show me the edit and wait for my OK.
Example input: "Quick update: leadership killed mobile and moved us to retention." That's a decision for the log, an edit to goals.md, and a risk for anything mobile-dependent.
Pitfall: letting the agent rewrite context files silently. Make it show the change and ask first.
Midweek: interview synthesis
What: themes, quotes and open questions from this week's customer calls. The Claude Code hub already has the full recipe with prompt and output: see the interview synthesis recipe on Claude Code for PMs. Run it on whatever transcripts landed this week and save to Work/research/.
Pitfall: synthesizing two interviews as if they were twenty. Ask the agent to say how many sources support each theme.
PRD day: draft, red-team, revise
What: a first draft grounded in this week's synthesis, then two critical passes before anyone else reads it.
Draft a PRD for <feature> from Work/research/<topic>-synthesis.md and
Context/. Then, in a fresh pass, list what the draft gets right. Then,
in another pass, find the weakest assumption and anything an engineer
would push back on. Revise and save to Work/prds/.
The PRD guide covers the full workflow and templates.
Pitfall: reviewing in the same pass that wrote the draft. Use a fresh pass or sub-agent so the reviewer isn't defending its own work.
Friday: stakeholder update and weekly review
What: one update for stakeholders, one honest review for you.
From this week's Notes/ and Work/decisions.md, write (1) a stakeholder
update: what moved, what's blocked and who can unblock it, decisions
made, next week's focus, under 200 words; and (2) a private review:
what I said I'd do vs what I did, and one thing to change next week.
Save both to Work/updates/.
Pitfall: an update that reads like a changelog. Lead with what you need from the reader.
Now and then: prototypes and competitor checks
When a PRD needs something clickable, use the prototype guide. For competitor checks, run the competitor monitoring recipe on the hub once a month rather than every day.