Guide · Agent skill
How to use the Product Pulse Reports agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
ce-product-pulse is an agent skill that pulls usage, performance, error and follow-up signals from your product's configured sources for a chosen time window, such as the last 24 hours or 7 days. It writes a compact report of roughly 30-40 lines and summarises the headlines in chat. This guide explains what agent skills are, how to install this one, how to ask for a report, and what it needs and touches before you run it.
Quick start
- 1Download the ZIP. Get the ce-product-pulse ZIP from this page.
- 2Unzip the skill. Unzip into ~/.claude/skills/ for personal use, or into .claude/skills/ in your project, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm your repository. Run the agent inside a Git repository so the skill can resolve the repo root.
- 4Complete the first-run interview. Answer the questions about your analytics, tracing and events so pulse_* keys are saved to config.local.yaml.
- 5Ask for a pulse. Request a report with a window such as 24h or 7d in plain language.
- 6Review the saved report. Open the dated file in pulse-reports/ and read the headlines and top followup.
What the ce-product-pulse skill makes the agent do
An agent skill is a folder of instructions and, here, reference files that an AI coding agent loads when a task matches it. Once this skill is installed, asking for a product pulse makes the agent query the data sources you have configured for a time window, then produce a single-page report.
The report covers usage, performance, errors and followups. It is saved to a pulse-reports folder inside your repository as a dated file named like YYYY-MM-DD_HH-MM.md, and the key points appear in chat. The skill presents numbers without labelling them good or bad, and it keeps personal data such as user emails, account IDs and message content out of the saved file.
- Reads from analytics, tracing and, optionally, a read-only database or MCP tools
- Saves a 30-40 line report under pulse-reports/ in the repo root
- Compares the current window with the previous window of equal length when possible
- Does no alerting and does not change your product, database or external systems
Install the skill from the ZIP
Download the ZIP for ce-product-pulse from this page, then choose one of two options. For a single personal setup, unzip it into ~/.claude/skills/. For a project that should share the skill with everyone who clones the repo, unzip it into the project's .claude/skills/ folder.
If you use the Claude apps instead of a local install, you can upload the skill under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into .claude/skills/ inside the repository
- Claude apps: Settings → Capabilities → Skills, then upload the skill
Set up the skill for your product
The first time you run it, the skill checks whether pulse_product_name is set in .compound-engineering/config.local.yaml or .compound-engineering/config.yaml. If it is not, the skill starts a first-run interview. The interview asks which tools you use and which events matter, and writes the answers as pulse_* keys into config.local.yaml, leaving your other keys alone.
If you use a database for this, it must be a read-only connection. The interview refuses read-write credentials. Database access is optional; many products get a useful pulse from analytics and tracing alone.
- Key settings include pulse_product_name, pulse_lookback_default (1h, 24h, 7d or 30d), pulse_analytics_source and pulse_tracing_source
- pulse_quality_scoring is off by default and is meant for AI products
Trigger a report with natural requests
Ask in plain language, for example for a daily check or a weekly summary. You can also pass a lookback window as the argument, such as 1h for a launch check, 24h or 7d. If you give no window, the skill uses pulse_lookback_default, and if that is unset, 24h.
The skill applies a 15-minute trailing buffer to the window's end, because analytics and tracing tools have ingestion lag. For a 24h window it queries from now minus 24 hours minus 15 minutes, up to now minus 15 minutes.
- "Run a product pulse for the last 24 hours"
- "Give me a 7-day product pulse report"
- "Do a 1h pulse, we just shipped the signup change"
- "Show me past product pulses" (reports are saved, so they browse as a timeline)
Read the report and its limits
The report opens with headlines, so the first three lines carry the most important change. Each figure is a real number from the source, with the source noted inline where it is uncertain. If a section has too little data, it stays thin rather than being padded.
The skill does not decide whether a number is good or bad, and it does not alert you. For deep investigation, use the native tools of your analytics or tracing provider. The pulse is a consolidated single-page read, not a dashboard.
- Headline lines come first
- Percent changes appear only when a comparison window is possible
- Shipped work is not tracked here; use your issue tracker and commit history
Safety notes: scripts and network access
This skill runs scripts and uses the network, because it queries your configured data sources. It needs a Git repository, since it resolves the repo root with git rev-parse --show-toplevel. Its tools include Read, Write, Glob, Grep, Bash and a blocking question tool.
Its writes are limited to the pulse settings in .compound-engineering/config.local.yaml and the report file. Data-source tools are used read-only; if a tool offers a write mode, the skill is designed not to use it. Do not give it read-write database credentials, which it is built to refuse.
- Requires a Git repository and configured data sources
- Network use is for querying your analytics, tracing or MCP sources
- Review the scripts and reference files in the ZIP before installing if your policy requires it
- Do not place personal data in the reports; the skill is designed to omit it
git rev-parse --show-toplevelWhat people use it for
- A founding engineer checking the last 24 hours of usage and errors each morning
- A small team reading a 7-day summary before its weekly product meeting
- A quick 1h check after a launch to see early usage and error signals
- Building a dated history of product pulses in the repo to look back at changes over time
Frequently asked questions
How do I install the ce-product-pulse skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into the project's .claude/skills/ folder for a shared setup. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
What does ce-product-pulse need to work?
It needs a Git repository, the agent tools Read, Write, Glob, Grep, Bash and a blocking question tool, and configured product data sources such as analytics or tracing. Database access is optional and must be read-only.
Is ce-product-pulse free to use?
The skill is published under the MIT licence. The source does not describe pricing for the analytics or tracing services you connect to, so check those providers directly.
Does the skill send alerts when something breaks?
No. The skill explicitly does no alerting. It presents numbers in a single-page report and leaves judgement and deep investigation to your native analytics or tracing tools.
Can it use a read-write database connection?
No. The interview refuses read-write credentials, because a write-capable connection would break the skill's read-only design. Use a read-only connection or skip database access.
Why does my report show a 15-minute gap at the end?
The skill applies a 15-minute trailing buffer to the window's upper bound because analytics and tracing tools have ingestion lag. Querying right up to now would under-report the most recent events.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use ce-product-pulse if you are a product or founding engineer, or part of a small team, that already has analytics or tracing and wants a regular, dated one-page snapshot saved in the repo. Pick something else if you need live alerts, deep investigation, or a full dashboard, or if you have no data sources to connect.
Download the skill