Claude Code Chief of Staff: Automate the Weekly Operating Rhythm
A chief of staff at a 300-person SaaS company spends most of Thursday and half of Friday doing the same thing: pinging eight team leads for updates, reading through Slack and project trackers, reformatting it all into one summary, and trying to spot which projects quietly slipped this week. By the time the operating review starts Monday, the work is stale and they’re exhausted. A claude code chief of staff workflow takes that assembly job and turns it into a standing session you run once a week, so you arrive at Monday with a draft already in hand and your judgment reserved for the parts that actually need it.
The point isn’t a clever one-off prompt. It’s a rhythm. The weekly operating cadence has four predictable moves, and each one is mostly mechanical until the moment a human has to decide something. That split is exactly where automation earns its keep.
The four moves of the weekly operating rhythm
Every functioning operating rhythm does the same four things, in order. Collect updates from every team. Flag what’s off track against the plan. Surface the decisions that need an owner this week. Close the loops from last week so nothing falls through.
Three of those four are assembly work. Reading twelve update threads and pulling out the status of each project is tedious, not hard. Comparing this week’s numbers to the committed plan and marking the gaps is rule-following. Checking whether last week’s open items got resolved is a lookup. Tools like Claude Cowork and Claude Code are good at exactly this kind of read-summarize-compare work when you give them consistent inputs and a clear instruction.
The fourth move, surfacing decisions, is where it gets interesting. Claude can flag that two teams disagree on a launch date or that a project has no owner. It cannot decide who’s right or who should own it. That’s your call, and it should stay your call.
Set up the standing session once
The trick is writing the instructions down once instead of re-prompting every Friday. Claude Code reads project files and a memory file, so you can describe your operating rhythm in plain language and have it run the same way every week. The memory documentation covers how a persistent instruction file works.
Create a single instruction file that names your inputs, your cadence, and your output format. Here’s a starting structure you can adapt.
# Weekly Operating Rhythm — Instructions
## Inputs (read these every run)
- /updates/*.md # team lead updates, one file per team
- /trackers/okr.md # quarterly commitments
- /summaries/last-week.md
## Output: weekly-operating-summary.md
1. Status by team — one line each, RAG color
2. OFF TRACK — anything behind the committed plan, with the gap
3. DECISIONS NEEDED — items with no owner, conflicts, blockers
4. LOOPS — open items from last week and whether they closed
Run it once against last week’s real data. The first output won’t be perfect. Maybe it over-flags minor slips as red, or buries a real conflict in a status line. Edit the instruction file, not the output, and run it again. Two or three cycles and it writes the summary the way you would. For a deeper template you can borrow from, the AI weekly planning template walks through the structure section by section.
Collect updates without the Thursday chase
The slowest part of the old way is gathering inputs from people who are busy. You won’t fully eliminate the ask, but you can shrink it. Standardize how teams drop updates so the same fields land in the same place each week. A short shared template per team, dropped into one folder, beats hunting through eight different Slack channels.
Once the inputs are consistent, the collect step becomes a single pass. Claude reads every update file, normalizes them into one status block, and notes any team that didn’t submit. That last part matters: knowing who’s missing is half the value, because the silent team is usually the one with the problem.
Flag off-track work against the plan
Flagging slippage is where consistency pays off. Give Claude the committed plan and the current status, and ask it to mark every gap with the size of the gap, not just a color. “Behind” is useless. “Behind by 9 days, no recovery plan noted” tells you where to spend Monday.
Keep the rule simple and explicit in your instruction file: anything that moved more than a set threshold from plan gets surfaced, with the delta. You’re encoding a judgment you’d otherwise make manually every week. Resist the urge to make the rules clever. Plain thresholds you can explain to your CEO beat a scoring model nobody trusts.
This is also where you decide what stays human. Claude can say a project is off track. Whether that’s a crisis or a non-event depends on context it doesn’t have, and you shouldn’t pretend otherwise. The honest version of this workflow flags, then hands you the read.
Surface decisions and close the loops
The two highest-value sections of your summary are decisions needed and open loops. For decisions, have Claude pull out every item where two sources disagree, where a project has no named owner, or where someone flagged a blocker they can’t clear alone. That becomes your Monday agenda, pre-built.
For loops, the workflow compares last week’s open items to this week’s status and reports what closed and what didn’t. Nothing dies quietly. This single step is what separates a real operating rhythm from a weekly status email that everyone ignores. If your reviews feed into board or leadership prep, the same outputs roll straight into board prep and recurring meeting prep instead of being rebuilt from scratch.
A note on trust: run the workflow as a draft generator, not an autosend. The summary that goes out should always pass through you first. You’re catching the rare miss, smoothing the framing on a hard message, and adding the one piece of context the documents never captured. That review is the job. Everything before it is the part you’re handing off. You can read more about how these standing sessions fit together on the chief of staff hub.
What this actually buys you
Done well, the weekly assembly drops from most of two days to a 30-minute review of a draft. You stop being the bottleneck on Friday and start the week with a clear-eyed view of what’s slipping and what needs a decision. The rhythm becomes reliable because it runs the same way regardless of how chaotic the week was.
What it doesn’t do is make the calls. The escalation, the framing, the read on whether a slip is noise or a signal, the hard conversation with the team lead who keeps missing commitments. Those stay with you, which is the point. You’re trading the clerical half of the role for more room to do the judgment half well. For the technical details of how Claude Code reads project files and runs repeatable workflows, the Claude Code overview is the place to start.
Frequently Asked Questions
What is a Claude Code chief of staff workflow?
It’s a standing, repeatable session where Claude reads your team’s updates, project docs, and trackers, then drafts the weekly operating summary a chief of staff normally assembles by hand. It collects status, flags off-track items, and surfaces decisions that need an owner. You review and edit before anything goes out.
Does this replace the chief of staff role?
No. It removes the assembly work, not the judgment. Deciding what to escalate, how to frame a hard message, and which fire to put out first stays human. The workflow gives you back the hours you’d otherwise spend chasing updates and reformatting them.
How long does it take to set up?
Most people get a usable first version in an afternoon. You write a reusable instruction file describing your cadence and where your inputs live, run it once against last week’s data, and refine the prompt over two or three cycles until the output matches what you’d write yourself.
What inputs does it need?
Whatever you already use to run the week: Slack threads, project trackers, meeting notes, OKR docs, and prior weekly summaries. The more consistent your source format, the cleaner the output. Claude works from what you give it, so connecting the same inputs each week is what makes the rhythm repeatable.
Is my company data safe in this workflow?
Treat it like any tool that touches internal docs: follow your company’s data policies and review what you connect. Keep sensitive personnel or legal matters out of automated summaries and handle those directly. Check Anthropic’s product documentation for current data handling details.
Pick one upcoming week, write the instruction file, and run it once against last week’s real updates before you trust it live. If you want a guided path to building standing sessions like this, the Cowork course walks through it end to end.