Current limitations
kipp•punkt is under active development. This page lists known boundaries and rough edges so you can plan around them.
No native multi-repo support
Section titled “No native multi-repo support”You can run kipp•punkt for multiple repositories on the same machine, but each instance polls GitHub independently. As you add repos, this becomes inefficient against GitHub’s API rate limits and harder to manage centrally. A centralized multi-repo mode is on the roadmap.
Limited observability
Section titled “Limited observability”When an agent run fails, diagnosing the root cause can be difficult. How much detail you get depends on your harness. Some surface useful logs, others don’t. There is no built-in dashboard or structured error reporting yet.
No CI pipeline reaction
Section titled “No CI pipeline reaction”The agent does not react to CI pipeline outcomes. If your build or test suite fails after a PR is opened, the agent won’t notice or attempt a fix on its own. You need to flag the failure in a review comment for the agent to act on it. For now, the best approach is to instruct the agent to run the relevant checks before committing, either in your AGENTS.md or via the pretext config property.
No persistent memory across tasks
Section titled “No persistent memory across tasks”Each task starts from a blank slate. The agent does not carry learned context, preferences, or corrections from previous tasks into the next one. If it made a mistake on one PR and you corrected it in review, it may repeat the same mistake on the next task. Persistent memory is on the roadmap.
Insufficient harness sandboxes
Section titled “Insufficient harness sandboxes”Current AI harness sandboxes are either too restrictive, not restrictive enough, or not supported at all. In practice, this means you should run kipp•punkt inside a Docker container to get a meaningful security boundary, or accept the risk of running the agent with full access to your machine.