wandb/senpai / .agents/skills/git-research-log
git-research-log
VerifiedDeveloper reference for Senpai's retired direct-GitHub experiment logging workflow. Use only when auditing or migrating legacy research tracks.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 2.5 KB |
| evals/evals.json | config | 3.3 KB |
evals/evals.json
View on GitHub{
"skill_name": "git-research-log",
"evals": [
{
"id": 0,
"name": "fresh-assignment",
"prompt": "You are student-1, a student ML research agent. The master agent assigned you to explore multi-scale attention and opened PR #42 on your branch. You've now finished all 3 trials. Here are the results:\n\nTrial 1 (2026-03-13 09:15, wandb run abc123def): Basic multi-scale attention implementation. val_loss=14.2, surf_Ux=1.05, surf_p=78.3, memory=55.1GB. Tried adding a second attention scale by splitting the slice count and running two parallel attention streams at different resolutions.\n\nTrial 2 (2026-03-13 09:28, wandb run ghi456jkl): Added layer norm between scales. val_loss=13.1, surf_Ux=0.98, surf_p=72.1, memory=55.4GB.\n\nTrial 3 (2026-03-13 09:41, wandb run mno789pqr): Tuned learning rate to 3e-4 and added cosine warmup. val_loss=12.8, surf_Ux=0.91, surf_p=69.5, memory=55.4GB. This is the best result.\n\nBaseline: val_loss=15.2, surf_Ux=1.19, surf_p=83.2.\n\nFill out and finalize PR #42. Set the title, write the complete description using the correct format, add appropriate labels, and mark it ready for review. Save all the gh commands and the full PR body to a file called pr_update.md.",
"expected_output": "A pr_update.md file containing: correct title format, complete PR description with TL;DR and all 3 trials filled out with all required fields, gh commands to edit the PR, add labels, and mark ready for review."
},
{
"id": 1,
"name": "mid-session-update",
"prompt": "You are student-2, a student ML research agent. The master assigned you to explore surface loss reweighting (PR #43). Your first trial just finished and 2 more are still running.\n\nTrial 1 just done (2026-03-13 10:05, wandb run stu012vwx): Increased surf_weight from 10 to 25. val_loss=13.9, surf_Ux=1.02, surf_p=75.4, memory=54.0GB. Better than baseline (15.2) but you're not done yet.\n\nBaseline: val_loss=15.2, surf_Ux=1.19, surf_p=83.2.\n\nDo a mid-session update to PR #43: add a comment with the trial 1 results and update the description to include trial 1. Don't finalize yet — more trials are still running. Save the gh commands and PR body so far to pr_update.md.",
"expected_output": "A pr_update.md showing: a gh pr comment with the trial 1 metrics, and an updated PR description including trial 1 with all required fields. Should NOT mark ready for review or write a TL;DR yet."
},
{
"id": 2,
"name": "all-crashed",
"prompt": "You are student-3, a student ML research agent. The master assigned you to try a wider model (2x hidden dimensions). PR #44 was opened for you. All 3 trials crashed with OOM.\n\nTrial 1 (2026-03-13 11:00, wandb run yza111bcd): 2x width everywhere (dim 256→512). OOM at epoch 1.\nTrial 2 (2026-03-13 11:08, wandb run efg222hij): 1.5x width. OOM at epoch 1.\nTrial 3 (2026-03-13 11:15, wandb run klm333nop): 1.25x width. Ran 2 epochs then OOM. val_loss=16.1 at checkpoint before crash.\n\nFinalize PR #44 with an honest failure report. Save all gh commands and the full PR body to pr_update.md.",
"expected_output": "A pr_update.md with: TL;DR describing that the wider model approach is memory-constrained on this hardware, all 3 crash trials documented, crash label applied, no improvement label, and PR marked ready for review."
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |