research-engineering
Skill4.1k starsChanged 49 days agoApache-2.0View on GitHub
Explore Marin designs, datasets, and experimental pipelines through iterative analysis and experiments; preserve a runnable research branch and notebook, then promote selected code into Marin PRs. Use for experimental programs, not bounded background research or routine debugging.
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use this skill when answering a question requires exploring alternatives, analyzing data or artifacts, or iterating on experiments before …
- Distinctive: line 13
Read the research engineering guide for the record, branch, and promotion workflow.
- Very specific: line 21
Start from the decision, baseline, constraints, and useful stopping condition.
- Very specific: line 22
Search prior work through use-echo and inspect the current checkout.
- Very specific: line 23
Use a task issue in marin-community/marin-logbooks as the public entry point.
- Distinctive: line 33
Keep the issue body current and its comments useful as a readable account of the important attempts, results, and changes of direction.
- Very specific: line 41
Use the domain skill for each concrete operation, such as add-dataset, write-pipeline, launch-rl, add-pallas-kernel, or profile-training.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Research engineering
Use this skill when answering a question requires exploring alternatives, analyzing data or artifacts, or iterating on experiments before choosing an implementation. Work can start before any run is launched. Background research alone follows experiment records.
Read the research engineering guide
for the record, branch, and promotion workflow. Read the current
marin-logbooks guidelines
before creating or updating its branches; research branches inherit Marin's
root instructions rather than the logbooks index branch's instructions.
Coordinate the work
- Start from the decision, baseline, constraints, and useful stopping condition.
Search prior work through
use-echoand inspect the current checkout. - Use a task issue in
marin-community/marin-logbooksas the public entry point. Reuse an existing issue; when the task authorizes a new public research program, usefile-issuewith target repositorymarin-community/marin-logbooksto create its coordinating issue. An existing Marin issue can link to it. Do not publish private work without authorization. - Work on a Marin-derived branch in logbooks. Preserve the notebook and supporting analysis alongside the actual experimental source changes.
- Record why each meaningful alternative was tried, the evidence it produced, what was learned, and the next decision. Commit and push supporting code and notebook entries before a milestone issue update or handoff.
- Keep the issue body current and its comments useful as a readable account of the important attempts, results, and changes of direction.
- Extract selected code into Marin contribution branches. Use
manage-prfor their validation, publication, and feedback; keep the research branch intact. A useful negative result can finish a program without a code PR.
Delegate execution
Use the domain skill for each concrete operation, such as add-dataset,
write-pipeline, launch-rl, add-pallas-kernel, or profile-training.
use-iris owns job operations and query-finelog owns telemetry queries.
This coordinating workflow does not authorize compute launches or recovery
actions beyond the task's scope. Record meaningful restarts under their actual
execution identities.
Babysit launched runs and PRs through a heartbeat: a scheduled follow-up in the
host application that resumes monitoring after the active turn. Keep one
monitor per task with its execution or PR identity, latest observation, owner,
and exit condition. Report meaningful changes and required action; stay quiet
while unchanged. Use use-iris for job completion checks and manage-pr for PR
feedback and exit conditions.
Trending
More talk lately.about the whole repository
Who's talking about marin-community/marin
- @pythonhub.dev“Marin Open-source framework for the research and development of foundation models. …”Bluesky · Jun 22, 2026 · 3 likes
- @sungkim.bsky.social“Project: marin.community Model: huggingface.co/marin-commun... Repo: github.com/marin-commun... Doc: …”Bluesky · Oct 30, 2025 · 3 likes
- Open-Athena/marin“Reproduce PlantCAD1 using Marin”GitHub · Sep 30, 2025 · 3 reactions
- openxla/xla“Fixed CTA-per-update assignment limits ragged-all-to-all device kernel throughput”GitHub · Aug 25, 2026 · 2 reactions
- minxin-zhg/namo“Add an explicit LICENSE file for this repository”GitHub · Feb 25, 2026 · 2 reactions
The biggest 5 of 55 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 55
Talked about on GitHub and Bluesky (55 times).
On GitHub · 50
- Sep 30, 2025Reproduce PlantCAD1 using Marin — Open-Athena/marin3 reactions · 10 comments
- Aug 25, 2026Fixed CTA-per-update assignment limits ragged-all-to-all device kernel throughput — openxla/xla2 reactions · 1 comment
- Feb 25, 2026Add an explicit LICENSE file for this repository — minxin-zhg/namo2 reactions · 1 comment
- Dec 6, 2025Add GCS support and Marin model evaluation — plantcad/plantcad-dev2 reactions
- Jul 4, 2026[category] Dataset curation & processing tooling — currentai-org/os-ai-map1 reaction · 1 comment
45 more on GitHub
- Jun 9, 2026exp: train a 1.5B model on contacts-v1 dataset — Open-Athena/MarinFold1 reaction · 2 comments
- Apr 27, 2026Compare enhancer data-curation strategies via gLM training + VEP eval — Open-Athena/marin-dna1 reaction · 5 comments
- Sep 9, 2025[core] Ray version check is too strict on projects which have a Ray dependency — ray-project/ray1 reaction · 4 comments
- Oct 28, 2026Improve Hero AgentX throughput on sixteen GB200s — marin-community/marin-logbooks0 reactions · 8 comments
- Oct 26, 2026[hero] Measure softmax collapse in pretraining checkpoints — marin-community/marin-logbooks0 reactions
- Oct 23, 2026[nightly] Supply CoreWeave storage keys to Iris — marin-community/vllm0 reactions · 1 comment
- Oct 23, 2026[hero] Record adjusted averaging and compact chat SFT — marin-community/marin-logbooks0 reactions · 1 comment
- Oct 23, 2026Preserve runnable Marin research branches alongside notebooks — marin-community/marin-logbooks0 reactions
- Oct 20, 2026Preserve exact token prefixes through task training — marin-community/MarinSkyRL0 reactions
- Oct 17, 2026Adopt Daytona credential and abandoned-response fixes — marin-community/MarinSkyRL0 reactions · 1 comment
- Oct 14, 2026marin-dupekit-native: Add version 0.1.11.dev38062449553 — riseproject-dev/python-wheels0 reactions
- Oct 13, 2026marin-iris-native: Add version 0.1.3 — riseproject-dev/python-wheels0 reactions
- Oct 13, 2026Train verified structure closure and continuation targets — marin-community/MarinSkyRL0 reactions · 1 comment
- Oct 11, 2026Separate Daytona credentials for training and evaluation — marin-community/MarinSkyRL0 reactions
- Oct 11, 2026Run installed Harbor agents through the shared rollout engine — marin-community/MarinSkyRL0 reactions
- Oct 11, 2026Add explicit connection cleanup to RecordProxy — marin-community/harbor0 reactions
- Oct 11, 2026Expose installed-agent request IDs before startup — marin-community/harbor0 reactions
- Oct 11, 2026Preserve masked DPO and task validation on the shared rollout engine — marin-community/MarinSkyRL0 reactions · 1 comment
- Oct 11, 2026Capture native agent tool histories with exact rollout tokens — marin-community/harbor0 reactions · 2 comments
- Oct 10, 2026Resolve published Marin dependencies for CatCount training — marin-community/MarinSkyRL0 reactions · 13 comments
- Oct 7, 2026skyrl: add an opt-in rollout-engine integration — marin-community/MarinSkyRL0 reactions · 2 comments
- Oct 5, 2026feat(watch): schedule scripts and agents through one queue — marin-community/loom0 reactions
- Oct 5, 2026Train d1536 Grug MoE models on DNA — Open-Athena/marin-dna0 reactions
- Oct 4, 2026Specify typed SkyRL author recipes and staged integration — marin-community/MarinSkyRL0 reactions
- Oct 3, 2026[mini-swe] Load batch defaults before context overrides — marin-community/harbor0 reactions
- Oct 3, 2026[mini-swe] Enforce hosted model context limits — marin-community/harbor0 reactions
- Oct 3, 2026[sotopia] Package the agent and classify evaluator failures — marin-community/harbor0 reactions
- Oct 2, 2026Preserve custom router selections in TPU GMM inference — marin-community/tpu-inference0 reactions
- Sep 23, 2026fix(loom): allow local deployment reconciliation — marin-community/loom0 reactions
- Sep 23, 2026Wait for queued trial trajectories before archiving samples — marin-community/harbor0 reactions
- Sep 23, 2026[RL] Add Hero Megatron training and replay support — marin-community/MarinSkyRL0 reactions · 1 comment
- Sep 23, 2026[release] Qualify immutable main-next GPU candidates — marin-community/vllm0 reactions · 1 comment
- Sep 20, 2026[iris] Use shared lease refresh for draft caches — marin-community/MarinSkyRL0 reactions · 1 comment
- Sep 17, 2026[RL] Preserve tokenizer portability in native HF exports — marin-community/MarinSkyRL0 reactions
- Sep 16, 2026[checkpoint] Bound S3 multipart uploads — marin-community/MarinSkyRL0 reactions · 1 comment
- Sep 12, 2026Support Nemotron Ultra RLVR blends — marin-community/MarinSkyRL0 reactions
- Sep 5, 2026Retain native ITL and reproducible inference benchmark evidence — marin-community/MarinSkyRLsays it tested it0 reactions · 2 comments
- Sep 3, 2026[metrics] Export embedded vLLM observability — marin-community/MarinSkyRL0 reactions
- Sep 2, 2026Explicit MPMD pipeline hangs with NCCL 2.30.7 and passes with 2.31.2 — NVIDIA/jaxpp0 reactions · 3 comments
- Sep 1, 2026[artifacts] Promote terminal results before completion hooks — marin-community/harbor0 reactions
- Sep 1, 2026mum publish: external links with #fragments are broken on htmlpreview.github.io — Open-Athena/mumwelt0 reactions · 1 comment
- Aug 27, 2026[Eval] Robust CPU-based Efficient MATH grader — marin-community/evalchemy0 reactions · 1 comment
- Aug 17, 2026fix(loom): trigger sessions from submitted PR reviews — marin-community/loom0 reactions
- Aug 17, 2026Make `experiments.coral` package in upstream marin available — Open-Athena/MarinFold0 reactions
- Aug 7, 2026[train] Grug EP=8 training runs a chronically unstable raw gradient norm — marin-community/MarinSkyRL0 reactions · 2 comments
On Bluesky · 5
- Jun 22, 2026Marin Open-source framework for the research and development of foundation models. https://github.com/marin-community/marin — @pythonhub.dev3 likes and reposts · 1 reply
- Oct 30, 2025Project: marin.community Model: huggingface.co/marin-commun... Repo: github.com/marin-commun... Doc: marin.readthedocs.io/en/latest/… — @sungkim.bsky.social3 likes and reposts
- Mar 18, 2026I have $0 hundred grand to give you, but PRs are welcome: github.com/marin-commun... — @al.merose.com1 likes and repost
- May 19, 2025You can read more in our: - Website: marin.community - GitHub: github.com/marin-commun... - Discord: discord.gg/J9CTk7pqcM - Documentation:… — @dlwh.bsky.social1 likes and repost · 1 reply
- Aug 25, 2026🚀 Skyrocketing! 🚀 (200+ new stars) 📦 marin-community / marin ⭐ 1,909 (+214) 🗒 Python Open-source framework for the research and… — @github-trending.bsky.social0 likes and reposts · 1 reply
In this repository28 other files it gives its agents.
- AGENTS.mdAGENTS.md
- experiments/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- ab-test-zephyrSkill
- add-datasetSkill
- add-grantSkill
- add-pallas-kernelSkill
- change-grugSkill
- curate-experimentsSkill
- file-issueSkill
- launch-rlSkill
- manage-hero-runSkill
- manage-prSkill
- marina-appSkill
- onboard-marinSkill
- profile-trainingSkill
- query-finelogSkill
- refresh-forkSkill
- review-grantSkill
- review-prSkill
- run-ferriesSkill
- simplify-codeSkill
- trace-pulumi-diffSkill
- upload-hf-modelSkill
- use-echoSkill
- use-fsutilSkill
- use-irisSkill
- write-pipelineSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

