crud-archive-run
Skill309 starsChanged 4 months agoApache-2.0View on GitHub
Durably ARCHIVE everything informative from a finished run / experiment before it's cleaned up or its cluster artifacts age out — ALL Harbor trace_jobs (raw per-trial traces), ALL ray logs, ALL stdout/stderr (incl. vLLM/serving logs), and wandb. Pack-rat by design: if it's potentially informative, keep it. Only skip the non-informative-or-huge (model weights/checkpoints, core/memory dumps, massive raw tmux-pane / terminal-recording bytes). Many-tiny-files → tar THEN rsync (never rsync thousands of small files raw). Use when concluding/archiving an experiment, before a cleanup skill `rm`s an on-disk tree, or before CoreWeave R2/pod artifacts get GC'd. Per-run-type component maps live below; WHERE each artifact lives per cluster is a pointer into `.agents/ops/<cluster>/` and `.agents/projects/{harbor,marinskyrl,ot-agent}/`.
In this repository50 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- analyze-datagen-campaign-summarySkill
- analyze-dataset-token-lengthSkill
- analyze-id-eval-rankingSkill
- analyze-job-history-irisSkill
- analyze-rl-behaviorSkill
- analyze-training-run-irisSkill
- build-tpu-image-irisSkill
- code-create-staged-planSkill
- code-execute-staged-planSkill
- commitSkill
- crud-otagent-supabaseSkill
- crud-purge-below-gate-evalsSkill
- crud-purge-stale-eval-placeholdersSkill
- datagen-create-task-datasetSkill
- datagen-job-cleanupSkill
- datagen-launch-irisSkill
- datagen-launchSkill
- datagen-reduce-dataset-snapshotsSkill
- datagen-standard-launchSkill
- debugSkill
- docs-deslopSkill
- eval-agentic-cleanupSkill
- eval-agentic-launch-irisSkill
- eval-agentic-launchSkill
- eval-standard-cleanupSkill
- eval-standard-launchSkill
- file-issueSkill
- monitor-cron-sweep-irisSkill
- monitor-cron-sweepSkill
- monitor-job-tablesSkill
- monitor-restore-irisSkill
- monitor-restoreSkill
- monitor-restore-unifiedSkill
- rl-agentic-job-cleanupSkill
- rl-agentic-launch-irisSkill
- rl-agentic-launch-jupiterSkill
- rl-job-health-deep-diveSkill
- rl-standard-job-cleanupSkill
- rl-standard-launch-leonardoSkill
- serve-model-vibe-testSkill
- sft-cleanup-hf-onlySkill
- sft-job-cleanupSkill
- sft-launchSkill
- supervisor-initSkill
- utils-cleanup-stale-sandboxesSkill
- utils-fix-permissionsSkill
- utils-reclaim-stale-snapshotsSkill
- write-testsSkill
- writing-styleSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 46
Keep large single logs (vllm.log) in the tarball or rsync them alongside.
- Very specific: line 67
Default: ~/Documents/experiments/<active|complete>/<exp>/run_archive/<run-id>/.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
crud-archive-run
Archive potentially informative artifacts; skip only files that are both non-informative and large.
✅ ARCHIVE (always — every run type)
- All raw Harbor traces — every per-trial dir under
trace_jobs/(RL) /eval_jobs/<name>/(eval) / the datagen trace dir:result.json,config.json,manifest.json,lock.json,agent/trajectory.json(the raw agent transcript — INFORMATIVE, keep even at multi-MB),verifier/(reward.txt,ctrf.json,test-stdout.txt),step_results,trajectory.summarization-*.json. - All ray logs — the ray session dir (raylet, gcs_server, per-worker
*.out/*.err,python-core-*). - All stdout / stderr — SLURM
.out/.err, the complete CoreWeave finelog,vllm.log,job.log, and per-trialtrial.log. - wandb — the local
wandb/run dir if present; else record the run URL/id in the archive'sMANIFEST. - Configs / launch command / rendered YAML / metric CSVs /
trainer_log.jsonl.
❌ SKIP (non-informative AND large)
- Model weights / checkpoints —
*.safetensors,*.pt,*.bin,global_step_*/, consolidated shards (they live on HF / R2; not useful for post-hoc debugging). - Core / memory dumps —
core.*,*.hprof, coredump trees. - Massive raw terminal-pane bytes —
*.pane(raw tmux pane dumps) andagent/recording.cast(asciinema) only when large and redundant withtrajectory.json; keep small casts. - Conda/uv/pip caches, extracted wheel trees,
__pycache__,.venv.
Mechanic — tar many-small-files, then rsync
On the source cluster/pod, tar the small-file tree before rsyncing it:
# on the cluster (SLURM) — one tarball per run, excluding the SKIP set
tar --exclude='*.safetensors' --exclude='*.pt' --exclude='*.bin' --exclude='global_step_*' \
--exclude='core.*' --exclude='*.pane' \
-czf /tmp/<run>_archive.tgz -C <run_dir> trace_jobs logs *.log config* wandb # adjust to what exists
rsync -aP <cluster>:/tmp/<run>_archive.tgz <dest>/ # then rm the /tmp tarballKeep large single logs (vllm.log) in the tarball or rsync them alongside. Verify the tarball is non-empty and
lists the expected trees (tar tzf … | head) before deleting the source.
Per-run-type components — WHAT + WHERE (pointers, they drift — read the ops/projects doc)
- CoreWeave agentic RL (SkyRL/MarinSkyRL) — durable traces:
s3://marin-us-east-02a/iris/<job>/trace_jobs(--trials-dir auto; pull withaws s3 --endpoint-url <R2>); pod-local traces:/app/experiments/<run>/trace_jobs(grab before pod GC withscripts/iris/analyze_coreweave_rl_job_live.sh <pod> cp). Full log:iris … job logs --since-ms <submit> --no-tail. Ray logs andvllm.logare pod-local; record the W&B URL. - SFT (LLaMA-Factory / axolotl, SLURM) —
.outper-step logs atexperiments/<job>/logs/*.out,trainer_log.jsonl, rendered config, wandb. Weights → HF (SKIP). Log path viascontrol show job <id> -oStdOut=/%Z. Details:.agents/projects/{llama-factory,axolotl}/, the cluster ops doc. - Datagen (Harbor traces) — one-level
trace_jobs/<trial>/result.json+ the harbor run log; the artifact is the trace set (→ HF), but archive the trace_jobs + logs. Details:.agents/projects/harbor/. - Eval (agentic Harbor) —
eval_jobs/<name>/<trial>/{result.json,config.json,agent/trajectory.json, verifier/,manifest.json,lock.json}+ top-levelvllm.log,job.log, per-trialtrial.log. Skip the bigrecording.cast/*.panewhen large. Details:.agents/projects/harbor/,eval-agentic-cleanup. - Cluster paths:
.agents/ops/iris/(CoreWeave),.agents/ops/tacc/(SLURM), and.agents/ops/empireai/(SLURM). Read the relevant one first.
Destination
Default: ~/Documents/experiments/<active|complete>/<exp>/run_archive/<run-id>/. Write a one-line MANIFEST
(run-id, cluster, job-id, dates, kept/skipped artifacts, W&B URL). Optionally push the tarball to HF
(penfever/…-archive, public default laion/) or R2. Archive and verify before any cleanup reclaims the tree.
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.

