Use Repster with AI coding agents¶
Repster is built CLI-first: pipelines are code, and every operation around them is an rp command, so an AI coding agent can operate the entire platform, from building and debugging flows to deploying and managing secrets, without a browser.
The agent dev loop¶
Everything an agent needs is plain text in, plain text out:
rp run my_flow # run the flow the agent just wrote
rp status # did it work?
rp logs # full output, including the complete traceback on failure
rp retry # re-run after the fix
Flows are ordinary Python in your repo, so the agent edits them like any other code and verifies locally with rp run, the same engine that runs in the cloud.
Load the shipped agent skill¶
uv add repster installs an agent skill alongside the package: a structured reference covering every command, concept, and guide. Find it:
python -c "import repster, pathlib; print(pathlib.Path(repster.__file__).parent / '.agents/skills/repster')"
Point your agent at the SKILL.md in that directory. For Claude Code, copy or symlink the directory into your project's .claude/skills/; other agents can read SKILL.md and its references/ directly.
Give your agent the docs¶
This documentation is published in AI-readable form:
- llms.txt: index of all pages
- llms-full.txt: the full docs in one file, for agent context windows
Scope keys for autonomous agents¶
For agents running unattended (CI, scheduled jobs), authenticate with an API key instead of an interactive login:
Creating API keys requires the org admin role; the key inherits that admin's role. See Manage API keys.