External readers see a healthy commit log and a 1.5.x release cadence, but no public sense of where the project is heading. This issue is a public tracking surface for post-1.5.2 candidates.
How to read this
Each bullet below is intent, not a commitment. If something gets picked up, it spawns its own issue (and ships in a future minor or patch). If a bullet sits here for a long time, that is also signal — comments and thumbs-ups help prioritize.
Candidates
-
GitHub Action: rk validate on PR
Drop-in workflow so any RK-governed repo can fail a PR when an agent leaves the project graph in a P0/P1 state.
-
Per-run cost / token telemetry
Capture token counts and dollar cost (where the adapter exposes them) into the run log, surface in rk run inspect.
-
First-class multi-agent races on a single sprint
rk lane parallel S-001 --agents claude,codex runs both, picks the better review verdict. Currently possible by hand; want it native.
-
VS Code extension
Sidebar showing rk runs, current review state, one-click rk close / rk discard. Reduces context-switching for users who live in editors.
-
Long-running agent observability
Stream agent stdout to a structured log so rk run logs --follow RUN-NNN works during long sessions, not just after.
How to influence priority
- Comment on this issue with your use case for any candidate.
- Open a new issue if you want to propose something not listed here.
- Pull requests welcome on items where the design is obvious — comment first to avoid duplicated work.
External readers see a healthy commit log and a 1.5.x release cadence, but no public sense of where the project is heading. This issue is a public tracking surface for post-1.5.2 candidates.
How to read this
Each bullet below is intent, not a commitment. If something gets picked up, it spawns its own issue (and ships in a future minor or patch). If a bullet sits here for a long time, that is also signal — comments and thumbs-ups help prioritize.
Candidates
GitHub Action:
rk validateon PRDrop-in workflow so any RK-governed repo can fail a PR when an agent leaves the project graph in a P0/P1 state.
Per-run cost / token telemetry
Capture token counts and dollar cost (where the adapter exposes them) into the run log, surface in
rk run inspect.First-class multi-agent races on a single sprint
rk lane parallel S-001 --agents claude,codexruns both, picks the better review verdict. Currently possible by hand; want it native.VS Code extension
Sidebar showing
rk runs, current review state, one-clickrk close/rk discard. Reduces context-switching for users who live in editors.Long-running agent observability
Stream agent stdout to a structured log so
rk run logs --follow RUN-NNNworks during long sessions, not just after.How to influence priority