experiment-bridge

Workflow 1.5: Experiment Bridge

Implement and deploy experiments from plan: $ARGUMENTS

Overview

This skill bridges Workflow 1 (idea discovery + method refinement) and Workflow 2 (auto review loop). It takes the experiment plan and turns it into running experiments with initial results.

Workflow 1 output:                    This skill:                                    Workflow 2 input:
refine-logs/EXPERIMENT_PLAN.md   →   implement → GPT-5.6-Sol review → deploy → collect → initial results ready
refine-logs/EXPERIMENT_TRACKER.md     code        (cross-model)    /run-experiment     for /auto-review-loop
refine-logs/FINAL_PROPOSAL.md

Constants

• CODE_REVIEW = true — GPT-5.6-Sol xhigh reviews experiment code before deployment. Catches logic bugs before wasting GPU hours. Set false to skip.
• AUTO_DEPLOY = true — Automatically deploy experiments after implementation + review. Set false to manually inspect code before deploying.
• SANITY_FIRST = true — Run the sanity-stage experiment first (smallest, fastest) before launching the rest. Catches setup bugs early.
• MAX_PARALLEL_RUNS = 4 — Maximum number of experiments to deploy in parallel (limited by available GPUs).
• BASE_REPO = false — GitHub repo URL to use as base codebase. When set, clone the repo first and implement experiments on top of it. When false (default), write code from scratch or reuse existing project files.
• COMPACT = false — When true, (1) read idea-stage/IDEA_CANDIDATES.md instead of full idea-stage/IDEA_REPORT.md if available, (2) append experiment results to EXPERIMENT_LOG.md after collection.

Override: /experiment-bridge "EXPERIMENT_PLAN.md" — compact: true, base repo: https://github.com/org/project

Inputs

This skill expects one or more of:

1. refine-logs/EXPERIMENT_PLAN.md (best) — claim-driven experiment roadmap from /experiment-plan
2. refine-logs/EXPERIMENT_TRACKER.md — run-by-run execution table
3. refine-logs/FINAL_PROPOSAL.md — method description for implementation context
4. idea-stage/IDEA_CANDIDATES.md — compact idea summary (preferred when COMPACT: true) (fall back to ./IDEA_CANDIDATES.md if not found)
5. idea-stage/IDEA_REPORT.md — full brainstorm output (fall back to ./IDEA_REPORT.md if not found)

If none exist, ask the user what experiments to implement.

Workflow

Phase 1: Parse the Experiment Plan

Read EXPERIMENT_PLAN.md and extract:

1. Run order and milestones — which experiments run first (sanity → baseline → main → ablation → polish)
2. For each experiment block:
   • Dataset / split / task
   • Compared systems and variants
   • Metrics to compute
   • Setup details (backbone, hyperparameters, seeds)
   • Success criterion
   • Priority (MUST-RUN vs NICE-TO-HAVE)
3. Compute budget — total estimated GPU-hours
4. Method details from FINAL_PROPOSAL.md — what exactly to implement

Present a brief summary:

📋 Experiment plan loaded:
- Milestones: [N] (sanity → baseline → main → ablation)
- Must-run experiments: [N]
- Nice-to-have: [N]
- Estimated GPU-hours: [X]

Proceeding to implementation.

Research-contract fallback: if idea-stage/docs/research_contract.md does not exist yet (idea selected outside /idea-discovery, or an older run), create it now from templates/RESEARCH_CONTRACT_TEMPLATE.md using the selected idea + claims from the experiment plan. Downstream /result-to-claim and /ablation-planner read this file as the claims source, and session recovery (docs/SESSION_RECOVERY_GUIDE.md) depends on it existing.

Phase 2: Implement Experiment Code

If BASE_REPO is set — clone the repo first:

git clone <BASE_REPO> base_repo/
# Read the repo's README, understand its structure, find entry points
# Implement experiments by modifying/extending this codebase

For each milestone (in order), write the experiment scripts:

1. Check existing code — scan the project (or cloned base_repo/) for existing experiment scripts, model code, data loaders. Reuse as much as possible.

2. Implement missing pieces:

   • Training scripts with proper argparse (all hyperparameters configurable)
   • Evaluation scripts computing the specified metrics
   • Data loading / preprocessing if needed
   • Baseline implementations if not already present
   • Fixed random seeds for reproducibility
   • Results saved to JSON/CSV for later analysis
   • Proper logging (wandb if configured in CLAUDE.md)

3. Follow the plan's run order — implement sanity-stage experiments first, then baselines, then main method, then ablations.

4. Self-review before deploying:

   • Are all hyperparameters from EXPERIMENT_PLAN.md reflected in argparse?
   • Is the random seed fixed and controllable?
   • Are results saved in a parseable format (JSON/CSV)?
   • Does the code match FINAL_PROPOSAL.md's method description?

Phase 2.5: Cross-Model Code Review (when CODE_REVIEW = true)

Skip this step if CODE_REVIEW is false.

Before deploying, send the experiment code to GPT-5.6-Sol xhigh for review:

mcp__codex__codex:
  model: gpt-5.6-sol
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    Review the following experiment implementation for correctness.

    ## Experiment Plan:
    [paste key sections from EXPERIMENT_PLAN.md]

    ## Method Description:
    [paste from FINAL_PROPOSAL.md]

    ## Implementation:
    [paste the experiment scripts]

    Check for:
    1. Does the code correctly implement the method described in the proposal?
    2. Are all hyperparameters from the plan reflected in the code?
    3. Are there any logic bugs (wrong loss function, incorrect data split, missing eval)?
    4. Is the evaluation metric computed correctly?
    5. **CRITICAL: Does evaluation use the dataset's actual ground truth labels — NOT another model's output as ground truth?** This is a common and severe bug.
    6. Any potential issues (OOM risk, numerical instability, missing seeds)?

    For each issue found, specify: CRITICAL / MAJOR / MINOR and the exact fix.

    === SCOPE LIMITS (these bound what you PROPOSE, never what you look for) ===
    Report anything that is actually wrong here — including a rare-looking case, if
    this repo actually produces it. Then keep the fix in scope:
    1. This is a RESEARCH-WORKFLOW tool, not a security paper. Verification is
       welcome; over-defense is not. Assume a cooperating operator on their own
       machine — a malicious local user is NOT in the threat model.
    2. Do NOT propose SHA / hash / content-fingerprint / digest-binding schemes.
       Reporting a real defect in hashing code that already exists is fine.
    3. NO speculative machinery: do not add feature flags, migration frameworks,
       compat layers, wrappers, pins, or similar mechanisms unless evidence shows
       a current repo defect they fix or an explicit existing invariant they must
       preserve. "Load-bearing", "compatibility", and "not scaffolding" are labels,
       not evidence. Point to the failing path/artifact or invariant, and check the
       proposal's factual premises, such as whether a named package version exists.
    4. NO corner-case obsession: exotic encodings, symlink races, RTL text and
       millisecond races are out of scope unless you can show the case arises here.
    5. Where a rubric or checklist is genuinely needed, do not over-mechanize
       judgement. A clear sentence a human reads beats a scored table nobody
       maintains.
    Exception: code that runs remote commands, starts a network service, or installs
    an MCP server runs on the user's machine with their credentials — trust-boundary
    findings there are in scope and the default is strict.
    Say plainly when something is correct. Do not manufacture findings.

On review results:

• No CRITICAL issues → proceed to Phase 3
• CRITICAL issues found → fix them, then re-submit for review (max 2 rounds)
• Codex MCP unavailable → skip silently, proceed to Phase 3 (graceful degradation)

Phase 3: Sanity Check (if SANITY_FIRST = true)

Before deploying the full experiment suite, run the sanity-stage experiment:

/run-experiment [sanity experiment command]

Wait for completion. Verify:

• Training loop runs without errors
• Metrics are computed and saved correctly
• GPU memory usage is within bounds
• Output format matches expectations

If sanity fails → auto-debug before giving up. Budget: up to 2 patch attempts on the same failure, then up to 2 clean reimplements (4 total):

1. Read the error — parse traceback, stderr, and log files. (The same read-the-primary-artifact discipline applies to surprising REVIEWER verdicts: see shared-references/review-tracing.md § Debugging With Traces.)
2. Diagnose — classify the failure:
   • OOM → reduce batch size or enable gradient checkpointing
   • ImportError → install missing package
   • FileNotFoundError → fix path or download data
   • CUDA error → check GPU availability, reduce model size
   • NaN/divergence → reduce learning rate, check data preprocessing
3. Fix and re-run — apply the fix, re-run sanity
4. Attempt 2+ still failing? → Call in Codex rescue (if Codex plugin installed): Before the next retry, invoke /codex:rescue to get a second opinion on the root cause. Codex independently reads the code and error logs — it may spot issues Claude missed (wrong tensor shapes, subtle import shadowing, config mismatches, etc.). Apply its suggested fix, then re-run.
   • If /codex:rescue is not available (plugin not installed), continue with Claude's own diagnosis
5. Both patch attempts failed on the same failure? → Discard and reimplement cleanly (up to 2 reimplements). Rewriting the failing script from EXPERIMENT_PLAN.md / the research contract is a PEER move to another patch, not a last resort — a third patch on top of two wrong ones is usually worse than a clean rebuild. Delete ONLY the attempt's own code/scaffolding (scripts this phase generated); the plan, EXPERIMENT_TRACKER.md, user-authored project source, collected data, and results are never deletable (see shared-references/external-cadence.md § Let a broken attempt restart, not just patch).
6. Budget exhausted (2 patches + 2 reimplements), or two reimplements failed the SAME way? → stop, report the failure with all attempted fixes and error logs. Two clean reimplements failing identically usually means the plan or the environment is wrong — say so explicitly in the report, because that (not the broken build itself) is what needs the human. Do not proceed with broken code.

Never give up on the first failure. Most experiment crashes are fixable without human intervention.

Phase 4: Deploy Full Experiments

Deploy experiments following the plan's milestone order. Route by job count:

Small batch (≤5 jobs per milestone) → use /run-experiment directly:

/run-experiment [experiment commands]

Large batch (≥10 jobs, multi-seed sweeps, or phase dependencies) → use /experiment-queue for proper orchestration:

/experiment-queue [grid spec or manifest]

Auto-routing rule: if any milestone in EXPERIMENT_PLAN.md declares ≥10 jobs (e.g., seeds: [42, 200, 201, ...] × N: [64, 128, 256] × n: [50K, 150K, 500K, 652K] = 36 jobs) or declares teacher→student phase dependencies, route that milestone to /experiment-queue. Otherwise use /run-experiment.

/experiment-queue adds: OOM-aware retry with backoff, stale-screen cleanup, wave-transition race prevention, phase dependency enforcement, crash-safe state persistence in queue_state.json. See skills/experiment-queue/SKILL.md for the manifest YAML format.

For each milestone:

1. Deploy experiments in parallel (up to MAX_PARALLEL_RUNS for /run-experiment, or max_parallel from manifest for /experiment-queue)
2. Use /monitor-experiment to track progress (reads from queue_state.json if /experiment-queue is active)
3. Collect results as experiments complete

🚦 Checkpoint (if AUTO_DEPLOY = false):

🔧 Code implementation complete. Ready to deploy:

Milestone 0 (sanity): [status — passed/pending]
Milestone 1 (baseline): [N experiments, ~X GPU-hours]
Milestone 2 (main method): [N experiments, ~X GPU-hours]
Milestone 3 (ablations): [N experiments, ~X GPU-hours]

Total estimated: ~X GPU-hours on [N] GPUs

Deploy now? Or review the code first?

Phase 5: Collect Initial Results

As experiments complete:

1. Parse output files (JSON/CSV/logs) for key metrics
2. Training quality check — if W&B data is available (CLAUDE.md has wandb: true and wandb_project), invoke /training-check to detect NaN, loss divergence, plateaus, or overfitting. If W&B is not configured, skip silently.
3. Update refine-logs/EXPERIMENT_TRACKER.md — fill in Status and Notes columns
4. Check success criteria from EXPERIMENT_PLAN.md — did each experiment meet its bar?
5. Write initial results summary:

# Initial Experiment Results

**Date**: [today]
**Plan**: refine-logs/EXPERIMENT_PLAN.md

## Results by Milestone

### M0: Sanity — PASSED
- [result]

### M1: Baselines
| Run | System | Key Metric | Status |
|-----|--------|-----------|--------|
| R001 | baseline_1 | X.XX | DONE |

### M2: Main Method
| Run | System | Key Metric | Status |
|-----|--------|-----------|--------|
| R003 | our_method | X.XX | DONE |

### M3: Ablations
...

## Summary
- [X/Y] must-run experiments completed
- Main result: [positive/negative/inconclusive]
- Ready for /auto-review-loop: [YES/NO]

## Next Step
→ /auto-review-loop "[topic]"

Phase 5.5: Write Compact Log (when COMPACT = true)

Skip entirely if COMPACT is false.

Append each completed experiment to EXPERIMENT_LOG.md:

## [Run ID] — [timestamp]
- **System**: [method name]
- **Config**: [key hyperparameters]
- **Result**: [primary metric = X.XX]
- **Verdict**: [positive / negative / inconclusive]
- **Reproduce**: `python train.py --config configs/run_id.yaml --seed 42`

This structured log survives session recovery — downstream skills read it instead of parsing screen output.

Phase 5.6: Auto Ablation Planning

After main experiments (M2) complete with positive results, invoke /ablation-planner to design ablation studies:

• Read the main results and method description
• Generate a claim-driven ablation plan: which components to remove, what to compare, expected outcomes
• Append ablation blocks to refine-logs/EXPERIMENT_PLAN.md and refine-logs/EXPERIMENT_TRACKER.md
• If main results are negative or inconclusive, skip ablation planning and note in the summary

If /ablation-planner is not available, skip silently — the existing EXPERIMENT_PLAN.md ablation blocks (if any) remain unchanged.

Phase 6: Handoff

Present final status:

🔬 Experiment bridge complete:
- Implemented: [N] experiment scripts
- Deployed: [N] experiments on [M] GPUs
- Completed: [X/Y] must-run, [A/B] nice-to-have
- Main result: [one sentence]

Results: refine-logs/EXPERIMENT_RESULTS.md
Tracker: refine-logs/EXPERIMENT_TRACKER.md

Ready for Workflow 2:
→ /auto-review-loop "[topic]"

Output Protocols

Follow these shared protocols for all output files:

• Output Versioning Protocol — write timestamped file first, then copy to fixed name
• Output Manifest Protocol — log every output to MANIFEST.md
• Output Language Protocol — respect the project's language setting

Key Rules

• CRITICAL — Evaluation must use dataset ground truth. When writing evaluation scripts, ALWAYS compare model predictions against the dataset's actual ground truth labels/targets — NEVER use another model's output as ground truth. Double-check: (1) ground truth comes from the dataset split, not from a baseline/backbone model, (2) evaluation metrics are computed against the same ground truth for all methods, (3) if the task has official eval scripts, use those.
• Follow the plan. Do not invent experiments not in EXPERIMENT_PLAN.md. If you think something is missing, note it but don't add it.
• Sanity first. Never deploy a full suite without verifying the sanity stage passes.
• Reuse existing code. Scan the project before writing new scripts. Extend, don't duplicate.
• Save everything as JSON/CSV. The auto-review-loop needs parseable results, not just terminal output.
• Update the tracker. EXPERIMENT_TRACKER.md should reflect real status after each run completes.
• Don't wait forever. If an experiment exceeds 2x its estimated time, flag it and move on to the next milestone.
• Budget awareness. Track GPU-hours against the plan's budget. Warn if approaching the limit.
• Vast.ai lifecycle. If using vast.ai instances, destroy them after all experiments complete and results are downloaded. Running instances cost money every second — don't leave them idle. Use /vast-gpu destroy or /vast-gpu destroy-all when done.
• Modal lifecycle. If using gpu: modal, no cleanup is needed — Modal auto-scales to zero after each run. But always show cost estimates before running and verify the spending limit is set at https://modal.com/settings (NEVER through CLI).

Composing with Other Skills

/idea-discovery "direction"          ← Workflow 1: find + refine + plan
/experiment-bridge                   ← you are here (Workflow 1.5: implement + deploy)
/auto-review-loop "topic"            ← Workflow 2: review + iterate
/paper-writing "NARRATIVE_REPORT.md" ← Workflow 3: write the paper

Or use /research-pipeline for the full end-to-end flow (includes this bridge).