assign-experiment

name	assign-experiment
description	Create a branch and draft PR to assign a hypothesis to a student. Handles branch creation, PR creation with the correct template and labels, and ensures the student gets a well-structured assignment. Use this skill to: assign an experiment, create a hypothesis PR, give a student work.
argument-hint	<student-name> <hypothesis-slug> <problem-dir>
model	claude-sonnet-4-6
effort	high

assign-experiment

Create a branch and draft PR that assigns a hypothesis to a student. The student will pick it up on their next poll cycle.

Arguments

• $0 — The student to assign (e.g. fern)
• $1 — A short kebab-case slug for the hypothesis (e.g. cosine-annealing)
• $2 — The active problem directory (e.g. target/cfd_tandemfoil)

The hypothesis details, instructions, and baseline metrics come from your own reasoning — this skill handles the git/GitHub mechanics.

Steps

1. Start from the latest advisor branch:

source "${CLAUDE_PLUGIN_ROOT}/scripts/senpai-gh.sh"
ADVISOR_BRANCH="${ADVISOR_BRANCH}"  # from environment

2. Create the experiment branch with a real assignment commit:

BRANCH="$0/$1"
create_assignment_branch "$0" "$1"

3. Write the PR body to a temp file. Replace the placeholders with your actual hypothesis, instructions, and baseline data:

BODY_FILE=$(mktemp)
cat > "$BODY_FILE" <<'PREOF'
## Hypothesis
<What we think will improve metrics and why. For non-trivial changes,
include links to papers or code that support the hypothesis.>

## Instructions
<Specific changes to make to $2/train.py — be concrete.
"Try a higher learning rate" is vague. Change lr from 5e-4 to 1e-3 and add cosine annealing with T_max=epochs" is actionable.>

## Baseline
<Current best metrics from BASELINE.md:
- val/loss: X.XXX
- Surface MAE metrics: p_in | p_oodc | p_tan | p_re
- Baseline W&B run: <run-id> (<wandb-link>)
- Reproduce command: `cd "$2" && python train.py ...`>
PREOF

4. Create and verify the draft PR:

create_assignment_pr_from_file \
    "$0" \
    "$BRANCH" \
    "<Hypothesis title — clear, specific, under 70 chars>" \
    "$BODY_FILE" \
    "$ADVISOR_BRANCH"
rm -f "$BODY_FILE"

Important details

• Read BASELINE.md before creating the PR take the most recent metrics from the file. The student needs concrete metrics to compare against.
• Do not skip create_assignment_branch. GitHub rejects PRs with no commits between the advisor branch and head.
• Pass the active problem dir as the third argument. Example: senpai:assign-experiment fern cosine-annealing target/cfd_tandemfoil.
• Be specific in instructions. The student implements exactly what you write. Vague instructions waste GPU time.
• Use --wandb_group in instructions when a hypothesis needs multiple iterations (e.g. "try surface weight 5, 10, 20") so related runs are grouped in W&B.
• One hypothesis per PR. Bundling multiple changes makes it impossible to attribute what worked.
• Use create_assignment_pr_from_file. It fails if the PR is not draft, targets the wrong base or head, or is missing $ADVISOR_BRANCH, student:$0, or status:wip.
• If the PR body is too long, keep the core info in the body file and add supplementary details as a follow-up comment.