Agent instructions
Contents
Agent instructions are standing instructions that PostHog cloud agents read at the start of a task run. They work like an AGENTS.md or CLAUDE.md file on your computer, but PostHog stores them and gives them to every cloud run that needs them.
Use them to tell agents how your team works. For example:
- Which repository owns which part of your product
- Which checks to run before a pull request
- How to write commit messages, pull request descriptions, and code comments
- Your personal preferences, such as the language or the level of detail in replies
The repository's own AGENTS.md, CLAUDE.md, and Cursor rules still apply. Use agent instructions for guidance that is not in a repository, or that applies to more than one repository.
Set your instructions
Go to Settings > AI > Agent instructions. The section has two fields:
| Field | Who can edit it | Where it applies |
|---|---|---|
| Project instructions | Project admins | Every cloud task run in the project |
| My instructions | Every project member, for their own text | Cloud task runs that you start |
Each field accepts up to 20,000 characters of Markdown. Agents read the project instructions first, then your personal instructions.
Which runs use them
| Run | Project instructions | My instructions |
|---|---|---|
| A run that a person starts, from PostHog AI, Slack, PostHog web, or a PostHog Desktop cloud run | Yes | Yes, from the person who started the run |
| Self-driving runs, such as scouts, report research, and implementation | Yes | No |
| Scheduled runs and workflows | Yes | No |
Local runs in PostHog Desktop do not use these settings. They read the instruction files on your computer.
Privacy and safety
- Anyone who continues a task that you started can see your personal instructions. Do not put secrets or private information in them.
- An agent in a cloud sandbox cannot change agent instructions. This stops one run from writing instructions for later runs.
- A change applies to runs that start after you save it.