artifact · oneshot category · $99 Ultra mint (gravy train) · $199 Cursor Ultra · SuperGrok Heavy ~$300 · Grok bot free · X Premium+

Verbatim /goal + steer (as oneshot)

Category scoring treats this as a single oneshot. The short steer below was a mid-run insert in practice — operator was lunching at their favourite Chinese restaurant and dispatched while eating — folded here as if it rode on the oneshot.

/goal read this research paper i did on previous art from myself. Then read my xbgst-kimi skills, the specialist lanes, the xbgst-stack plugin. We are first creating, updating and rebasing the core runtime and core deps, to be initially ported from last known good, with the research paper from idea conceptualization, anti patterns, technical evidence and also preferred methods of execution. The current art conceptual framework is the one outlined in veigapunk.github.io/charter-ufo-fsd

Less prio but also foundational: Make sure that this orchestrator is not this garbage ass google gemini model for this run. 

Also, if possible orchestrate a self iterating loop that does it all e2e and tests locally, empirically, uses the core skill pack of godspeed, the wwkd planner posture before the judge orchestrated rounds, and dispatches multiple rounds of self iteration. Planners should be dispatched following the same orchestration framework we have going on. Specialist lanes should be the ones i have outlined in the related repos. Implementation, testing, self iteration, adaptabilty mid run and portability to other substrates and languages is not core at init but get increasingly important as self iteration goes on. Ideally we are able to serve the best possible version for each of the clis outlined in charter-ufo-fsd, allow for heavy parallelization and keep the essence but adapt the shape so each target substrate have a proper package to be able to work on it itself (outside of this cloud env, locally, as all things should be).

self handoff and while doing the iteration, auto fork your own work, when justified, so we can also adapt deliverables to other substrates listed. [make sure to highlight this is ufo-fsd in action.]

also, this rerouting we just did is actually something we should implement; task it to understand the blockage and do just like we did, the L1 swarm orch reroutes models to next best option, given available options (ufo-fsd conceptual framework) [we have done it manually, i had to check that models were usage limited and task you to re awake them using the other models bucket | ideally, this can be done naturally within the environment, with no required input outsourcing to the operator]

Manual recovery that worked (encode as orch capability):
- Detect ERROR / rate-limit park on swarm lanes (prompt_group peers).
- Allowed recovery bucket: Cursor Grok 4.6 fast (cursor-grok-4.6-high-fast) | Cursor Grok 4.5 fast (cursor-grok-4.5-high-fast) | Composer 2.5 (composer-2.5).
- Wake + AddAsyncFollowup with modelDetails.modelName = next-best from bucket (keep if already allowed; else round-robin). Reroute existing bcIds — do not mint new doom-loop agents.
- L1 swarm orch owns this loop natively (charter-ufo-fsd): no operator paste required once shipped.

UFO-core cloud oneshot · prompt-group swarm · submit_as: oneshot · $99 Ultra mint (gravy train) · $199 Cursor Ultra · SuperGrok Heavy ~$300 grant · Grok bot free · X Premium+