Most of what gets sold as AI agents for franchise operations is a dashboard with a sales team. This is not that. It is a working agent you install into the Claude or ChatGPT account you already pay for, from one click or one pasted line, and run on this week's actual numbers, in whatever messy mix of pasted rows and manager emails they arrived. We build agents like this for companies with no engineers on payroll; this one we publish free, because the weekly roll-up is where a franchise ops lead's Monday goes to die and it is the easiest place to prove what an agent honestly does.
For operators · free agent
Free AI agent for franchise operations
Turns a messy pile of per-location numbers into one weekly ops summary with exception flags, and never guesses a figure.
- Paste every location's numbers in whatever shape they arrive and get one weekly summary: the per-location table, exception flags, and the three questions for the ops call.
- It recomputes every total and percentage your reports claim, and says out loud when a location's math does not match its own figures.
- A silent location shows up as a flagged NOT SENT row, so a missing report reads as an exception instead of quietly disappearing.
- Installs into ChatGPT or Claude from one click or one pasted line: no download, no email gate, and nothing to configure.
Built by Maya Thompson at agentclaw
Install on ChatGPT
opens ChatGPT and sends the instruction
copies the instruction, opens Claude
Or paste it yourself, Claude Code included
Create this skill for me, exactly as its files specify: https://www.agentclawhq.com/agents/ops-reporting-agent-for-franchise-operations/skill.md- Version
- v1.0
- Published
- Aug 13, 2026
- Works with
- ChatGPT · Claude · Claude Code
- Price
- Free, no email gate
What it takes off your desk
01
Builds the weekly roll-up from pasted spreadsheets, POS exports and manager emails: one table, system totals marked computed, gaps marked NOT SENT.
02
Flags exceptions: silent locations, double-digit swings, labor over your stated target, safety complaints and cash discrepancies, with your own thresholds replacing the defaults when you give them.
03
Recomputes the totals, percentages and week-over-week changes your reports claim, and shows both figures whenever the math disagrees.
04
Reads a struggling location's trend and separates what the numbers show from hypotheses that need a conversation to confirm.
05
Drafts the weekly all-locations note, the one-to-one check-in email, and a thirty-minute ops call agenda from the same summary.
What a run looks like
example run
you paste
Copper Kettle Coffee, week ending Aug 9, 2026. Our labor target is 28%. Numbers below, Maple Grove's update came by email, still nothing from Riverside. Location | Net sales | Last week | Labor % | Complaints Downtown | 31,450 | 30,200 | 26.1% | 1 (slow service Tue) Northgate | 18,900 | 21,480 | 31.4% | 3 (one says lid failed and coffee spilled on their hand, other two are wait times) Old Mill | 24,120 | 23,900 | 25.2% | (didn't report) Cedar Falls | 12,300 | 12,050 | 27.8% | 0 Email from Maple Grove (Dana): "Week was solid, we did 22,400 which is up 8% on last week's 21,300. Ran about 480 labor hours. No complaints logged. Espresso machine was down Tuesday morning, tech came Thursday."
the agent returns
WEEKLY OPS SUMMARY: Copper Kettle Coffee, week ending Aug 9, 2026
Labor target used for flags: 28% (as stated).
BY LOCATION
Location Net Sales WoW Labor % Complaints
Downtown $31,450 +4.1% 26.1% 1 (slow service, Tue)
Northgate $18,900 -12.0% 31.4% 3 (1 safety: lid failed;
2 wait times)
Old Mill $24,120 +0.9% 25.2% NOT SENT
Cedar Falls $12,300 +2.1% 27.8% 0
Maple Grove $22,400 +5.2%* NOT SENT 0 (none logged)
Riverside NOT SENT NOT SENT NOT SENT NOT SENT
SYSTEM $109,170 +0.2% 27.6% 4
* Maple Grove WoW shown as computed from the dollar figures given (Dana
reported +8%; see Arithmetic Checks).
SYSTEM net sales and WoW cover the 5 locations with numbers (Riverside
NOT SENT). SYSTEM labor % is an unweighted average of the 4 locations
that reported a percent (Maple Grove gave hours, not a percent;
Riverside NOT SENT). SYSTEM complaints total covers the 4 locations
that reported a count (Old Mill and Riverside NOT SENT).
ARITHMETIC CHECKS
Maple Grove (Dana) reports net sales up 8% on last week's $21,300. The
figures given ($22,400 vs $21,300) compute to a 5.2% increase, not 8%.
EXCEPTIONS
- Riverside: no report received this week, every metric NOT SENT.
- Northgate: one of three complaints alleges a safety issue, a lid failed
and coffee spilled on a customer's hand.
- Northgate: net sales down 12.0% week over week, labor 3.4 points above
the 28% target.
- Maple Grove: claimed WoW of +8%, the inputs compute to +5.2% (see
Arithmetic Checks above).
THREE QUESTIONS FOR THE OPS CALL
1. Northgate: sales down 12.0% and labor at 31.4% in the same week, plus
the lid failure. Is this one bad week or a pattern, and has the lid or
cup supplier been checked?
2. Maple Grove: Dana's 8% growth claim computes to 5.2% from the numbers
she gave. Which figure is right, and did Tuesday's espresso machine
outage hold sales back?
3. Riverside: still no report. What's blocking it, and who is following
up this week?
[excerpt of the full summary: the missing-numbers list is in the complete
output]How installing works
Every path does the same thing: it hands your assistant one instruction pointing at this agent's source file, and the assistant reads the files and builds the skill in your own account. The ChatGPT button arrives with the instruction already submitted. The Claude button copies it to your clipboard and opens a new chat: paste and send. In Claude Code, or anywhere else, paste the same line from the panel. There is nothing to download and no account with us. If browsing is switched off on your plan, open the source link in your own browser instead: it is plain text, and copying it into a chat does the same job.
You need
- A ChatGPT or Claude account that can browse the web, which is how the one-line install fetches the agent's files.
- Your locations' numbers in any text form: pasted rows, a POS export, or the manager's email.
The details
Who this is for
A franchisor's ops lead with five to fifty locations, or a multi-unit owner wearing the ops hat on top of two others. Location reports arrive every week in three formats and four moods: a spreadsheet from one manager, an email from another, silence from a third. Somebody turns that into the Monday picture, and right now that somebody is you, an hour before the ops call.
If you run a system big enough to have analysts and a BI stack, this agent will still read a week's reports faster than a person, but it was not built for you. It was built for the operator whose reporting stack is a group chat and a spreadsheet named FINAL_v3.
How it fits your week
Numbers land Friday through Sunday. You paste the pile in as it stands: rows, exports, Dana's email, all of it at once. The agent builds one summary in the same shape every week: the per-location table with the metrics your system actually tracks, system totals marked as computed, and week-over-week only where a prior week genuinely exists.
Then it flags what needs you. A location that sent nothing appears as a flagged NOT SENT row rather than vanishing. Swings past ten percent get called out in either direction, because good surprises deserve checking too. Labor over the target you stated, safety complaints, cash discrepancies: flagged, always. And when a manager's own claimed percentage does not match their own figures, the agent shows both numbers and says so plainly, which is a conversation you want to have Monday rather than at year-end.
From the same summary it drafts the pieces that follow: the all-locations note that never shames anyone by name, the one-to-one check-in email that treats a franchisee as a partner having a hard month, and a thirty-minute agenda for the ops call.
What it deliberately does not do
It does not invent numbers. A missing metric is NOT SENT and lands on the missing-numbers list; an estimate would be fiction wearing a decimal point.
It does not import industry benchmarks. Your locations get compared to each other, to their own history, and to the targets you set, because those are the only numbers that are actually yours. General figures come out only if you ask, labeled as general.
And it does not play lawyer. Royalty disputes, territory rights, termination clauses: it names those as your attorney's territory in one sentence and gets back to the reporting. It will compute a royalty only from a formula you state, never from an assumption about what your agreement probably says.
What is inside the agent
Four plain-text files, and you can read every one of them before you trust them: the install source link on this page serves them as they are. SKILL.md is the agent itself: the four jobs, the hard rules, the voice. references/metrics-checklist.md is what it extracts from every report and the default flag thresholds, which your own thresholds replace the moment you state them. references/templates.md holds the exact output shapes, so this Monday's summary looks like last Monday's. README.md repeats the install instruction offline. No installer, no telemetry, no account with us. It is the same file format we use on client work, packaged as a Claude Skill, and versioned, so when we improve it the page shows a new number and reinstalling takes a minute.
Fair questions
Is the example on this page real?+
Yes, with one honest caveat. Copper Kettle Coffee is a sample system we wrote for testing, and the output is an unedited excerpt of the agent's actual run on it, including the arithmetic catch. We do not publish anyone's real location numbers, for the reason you would hope.
Does it connect to my POS or accounting system?+
No, and it does not pretend to. You paste text: exported rows, report emails, whatever your managers send. That is the honest trade of an installable agent against a platform, and it is also why setup takes minutes instead of a quarter.
I am a franchisee, not the franchisor. Is this still for me?+
Yes. A multi-unit owner runs the same Monday: several locations, mixed reporting, one call. The agent also keeps one location's numbers out of another's check-in email, which matters in both directions.
What happens when a manager's numbers are wrong?+
The agent recomputes what the report claims and shows both figures when they disagree, without silently correcting either. You decide which is right; it just refuses to let the mismatch pass unremarked.
What happens to our numbers?+
They go where you paste them: your own Claude or ChatGPT account, under that platform's terms. Nothing reaches us, ever; there is no account with us to reach.
Why is a paid agency giving this away?+
Because the library is our proof of work. Some systems install an agent and run it themselves forever, which is a fine outcome. Some decide they want agents built around their own operations and run for them month after month, and that service is the business we are in.
More agents like this
Want it built around your workflow, and run for you?
This agent is the generic version. We also build agents around the way your company actually works, install them, and run them for you month after month.