weekly-report

---
name: weekly-report
description: Turns last week's raw numbers into a one-page weekly business review showing what moved, what is off target, and who is fixing it by when. Use when someone asks for a weekly report, a Monday update, a KPI summary or a status page.
---

# The one page that tells you how the week went

A box score of every metric, plus a short exception list naming one owner and one date per problem.

## Claude does

1. Get four inputs before starting anything: the data export (one row per metric per period, dated), the metric list with a named owner each, the targets, the week-ending date.
2. Tag every metric **input** (what the team controls: calls made, hours booked, stock counted) or **output** (revenue, margin, cash). Inputs first. Outputs are the score, inputs are the steering wheel.
3. Cap the page at 47 metrics: Amazon reads 400 to 500 in 60 minutes, the simplified version 47 in under 15. Offered more, list them and ask which to cut. Never cut silently.
4. Specify the 6-12 view: trailing 6 weeks left, trailing 12 months right, one shared X-axis, last year faded behind this year, a target marker where a target exists. Six weeks, not four, so a month boundary cannot distort the trend.
5. Comment on exceptions only: off target, or a reversed trend. Nothing gets commentary for being fine.
6. Give every exception one named person, one deliverable, one date. Never a team, never "someone should".
7. Refuse to guess. No filled-in weeks, no annualised part-months, no unevidenced cause. Write "cause not known" and leave it.

## Then Codex does

Codex takes the bulk arithmetic: parsing the export, building each metric's trailing-6-week and trailing-12-month series, aligning last year, computing variance to target, rendering the charts and box score. One calculation repeated across every row, where people make copying errors.

Fill in {DATA_FILE}, {METRIC_LIST}, {TARGETS}, {WEEK_ENDING} and {OUTPUT_FOLDER} inside CODEX.md first, then run:

```
codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
```

## Claude checks

1. Output metric count equals the count in {METRIC_LIST}. None invented, none dropped.
2. Every chart shows 6 weeks and 12 months on a shared axis, prior year present.
3. Re-read three figures **from the source file**, not from the report. Never retype a number into a table and then reason from it: that habit once turned a correct ledger into an apparent 99% undercharging fault.
4. Every gap reads "no data". Nothing interpolated or zero-filled.
5. Every exception row carries owner and date.
6. No commentary against an on-target metric.

Any failure: name the failing rows and rerun Codex. Never hand-patch the output.

## Rules

- Exception-driven. Silence means on target.
- "Cause not known" beats a plausible story. Never explain a movement you cannot evidence.
- Six weeks, never four. Inputs before outputs.
- One owner, one deliverable, one date.
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is.

## If Codex is not installed

Claude does the parsing, maths and charts itself and says in the reply that it did.

## Built from

- Commoncog, "The Amazon Weekly Business Review", https://commoncog.com/the-amazon-weekly-business-review/ - fullest public account of the WBR: 6-12 graphs, faded prior-year line, box scores, 47 metrics in 15 minutes.
- w95/awesome-claude-corporate-skills, https://github.com/w95/awesome-claude-corporate-skills - 189 stars from api.github.com; source of 07-operations/project-status-report.
- ComposioHQ/awesome-claude-skills, https://github.com/ComposioHQ/awesome-claude-skills - 74,501 stars from api.github.com; internal-comms prior art.
- Anthropic, "Skill authoring best practices", https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices - official SKILL.md spec.

# The one page that tells you how the week went - Codex task

## You are given

- **{DATA_FILE}** - a CSV or spreadsheet export, one row per metric per period, with columns for date, metric name and value. It may contain metrics you do not need and periods older than you need.
- **{METRIC_LIST}** - the metrics to report, in page order, each tagged `input` or `output` and each carrying one named owner.
- **{TARGETS}** - a target per metric. Some will be blank.
- **{WEEK_ENDING}** - the last date of the reported week, ISO format.
- **{OUTPUT_FOLDER}** - where to write everything.

## Produce

1. `weekly-report.md` - a box-score table, one row per metric in {METRIC_LIST} order, with columns: metric, input/output, this week, prior week, 6-week average, same week last year, month to date, trailing 12 months, target, variance to target (absolute and per cent), owner, status (`on target` / `off target` / `no data`).
2. `charts/<metric>.svg` - one file per metric. Two panels on one shared X-axis: left panel the trailing 6 weeks ending {WEEK_ENDING}, right panel the trailing 12 months. This year's line solid; the same series one year earlier faded behind it; a horizontal target marker where {TARGETS} supplies one.
3. `exceptions.csv` - one row for every metric that is off target or whose 6-week trend reversed direction. Columns: metric, this week, target, variance, owner, cause. **Leave `cause` empty.**
4. `data-gaps.txt` - every metric and period with no value in {DATA_FILE}.

## Rules

- Every figure is computed from {DATA_FILE}. Never estimate, interpolate, smooth or carry a value forward.
- A missing period is a visible gap in the chart and a line in `data-gaps.txt`. Never a zero.
- Six weeks, not four. Twelve months, not four quarters.
- Never drop a metric because its data is patchy. Include it with status `no data`.
- Never add a metric that is not in {METRIC_LIST}.
- Write no commentary, cause or explanation anywhere. That is not your half of the job.
- Percentages to one decimal place. Keep the source file's currency and never round a source value.
- If a column name in {DATA_FILE} does not match what you expect, stop and report it rather than guessing the mapping.

## Return

Print to stdout:

- the number of metrics processed and the number in {METRIC_LIST}; they must match,
- every file written, with its byte size,
- the first three rows of the box score, verbatim,
- every row of `exceptions.csv`,
- the full contents of `data-gaps.txt`,
- any row of {DATA_FILE} you could not parse, verbatim.