Investigate Marin questions or coordinate experiments across sessions by reviewing prior work, testing explanations against conflicting evidence, and recording conclusions. Results include source links, limitations, and confidence in the conversation or an existing project record.
Research an explicitly requested Marin question or coordinate a multi-session experiment; find prior work, assess evidence, and record results in the task's existing durable surface.
Review the instructions the agent will receive before you install.
Use this skill for a requested prior-work brief or research program. Ordinary implementation does not require a research workflow.
Return a compact brief in the conversation for bounded work. For durable or cross-session work, use the task's existing issue, PR, W&B run or report, published report, or durable session channel. Do not create standalone repository design, research, plan, logbook, or snapshot files. If the user explicitly requests a particular artifact, honor that request.
Update docs/ or the relevant OPS.md when the result changes reusable product
or operational guidance. Keep raw logs and dense data in their source systems.
For experiment results, record the exact command, source revision, material configuration, hardware, baseline, result, and interpretation. Label estimates and exploratory results. Record failures that rule out a hypothesis; omit routine debugging history.
Use W&B for scalar series, plots, large comparison tables, or raw artifacts that are too dense for the narrative record. Runs that require direct comparison must share a project. Verify row counts, key uniqueness, aggregation, and the numbers cited in the narrative before publishing a claim.
Scale the response to the decision. Include:
Do not require a special branch, experiment ID, hypothesis queue, update cadence, issue, tag, or W&B project unless the task itself needs one.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Compare Zephyr changes using the same prepared dataset and matching execution settings. Analyze Finelog metrics for each processing stage to distinguish CPU efficiency, elapsed time, and peak memory changes, then report improvements, regressions, or tradeoffs.
Register Hugging Face datasets for Marin by inspecting their structure without downloading the full dataset. Identify text fields, configurations, and data splits, then add a dataset module and validate the mapping with a sample.
Turn a specified GCP access request into a reviewable Pulumi change and pull request. Select the narrowest suitable role and encrypt personal email identities, covering resource permissions and Cloud Run viewer access through IAP, an access gateway.
Add or update Pallas/Mosaic kernels, specialized computation routines for JAX. The skill covers reference implementations, output and gradient checks, public wrappers, fallback behavior, performance benchmarks, and hardware-specific tuning when requested.
Preserve old experiment scripts with a dated Git tag, a named snapshot of the code, and add comments linking related GitHub issues to the archived files and cleanup pull request.
Launch and continuously monitor a specified Zephyr pipeline on Iris when asked. It checks coordinator and worker health, tracks stage progress, identifies stalled or repeated failures, and requires explicit approval before stopping or restarting jobs.