wandb/senpai / .agents/skills/bootstrap-target
bootstrap-target
VerifiedCreate or improve a Senpai target repository's program.md. Use this skill whenever the user wants to point Senpai at a fresh ML or research target repository, define the research objective, primary metric, benchmark contract, allowed edit boundaries, W&B reporting contract, or prepare a repo for autonomous advisor/student experiment loops.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.6 KB |
| references/benchmark-integrity-patterns.md | markdown | 3.1 KB |
| references/interview-question-bank.md | markdown | 2.6 KB |
| references/program-template.md | markdown | 6.0 KB |
references/interview-question-bank.md
View on GitHub# Interview Question Bank
Ask only the questions needed to resolve durable ambiguity. Prefer questions
that expose tradeoffs over questions that ask the user to summarize files you
can read.
## Success And Research Strategy
- What does success look like?
- What would make this research run feel like a clear win rather than a modest
cleanup?
- What balance should Senpai strike between tuning known-good ideas and taking
bigger research bets?
- Are there related fields, domains, papers, benchmarks, or research
communities the researcher agent should draw inspiration from when proposing
experiments?
- Are there approaches you already believe are promising, over-explored, or
off-limits?
## Metrics
- What exactly should Senpai optimize?
- What is the primary metric, exact logged key/name, and direction?
- Is that metric already measured by the repo? If not, where should it be
added?
- What validation metric selects checkpoints?
- What test metric supports final claims?
- Are there secondary metrics that must not regress, even if the primary score
improves?
- Is there a statistical rule, seed policy, or confidence threshold for final
claims?
## Benchmark Integrity
- Which files define the split, hidden labels, evaluator, or benchmark contract?
- What changes would invalidate a result even if the metric improves?
- Are students allowed to change the model architecture, data processing,
optimizer, batch size, number of steps, dependencies, or evaluation code?
- Are any external sources, upstream repos, papers, branches, or post-launch
updates banned during the run?
- Is validation peeking, seed selection, early stopping, or adaptive trial
selection allowed?
## Data And Domain Context
- Where does data live in Senpai pods?
- What are the input and target shapes?
- What do feature and target channels mean?
- Which splits test in-distribution performance, OOD generalization, or
paper-facing claims?
- Are there known hard regimes, rare cases, domain constraints, or failure
modes that advisors should account for?
## Operations
- What command should students run for debug, screening, and confirmation runs?
- What GPU count, wall-clock timeout, epoch/step budget, and memory limit apply?
- What W&B entity/project/group naming should be used?
- What baseline should new PRs compare against?
- What result fields must appear in student PR comments?
## File Boundaries
- Which files are primary editable entrypoints?
- Which files are read-only context?
- Which files are protected and should never be touched in normal experiment
PRs?
- Are dependency changes allowed? If yes, where should they be recorded?
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |