academic-experiments
Skill111 starsChanged 5 months agoMITView on GitHub
Audit, run, or verify experimental evidence for CS/AI/ML papers. Produces Evidence Inventory with evidence_type annotations (newly_run/preexisting_artifact/user_claim) and Protocol Risk assessments. Use when: checking if experiment results are reproducible, auditing existing experiment artifacts, running minimal reproducible commands, evaluating checkpoints without full retraining, documenting protocol risks like data leakage or missing baselines. Triggers on: 复核实验, run experiments, 实验结果, experiment evidence, verify results, 实验验证, evidence inventory, protocol risk, 跑实验, check results, reproduce experiments, 实验审计.
In this repository7 other files it gives its agents.
- academic-citationSkill
- academic-figureSkill
- academic-latex-layoutSkill
- academic-paper-writerSkill
- academic-polishingSkill
- academic-reviserSkill
- academic-venue-researchSkill
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 12
Read manifest.yaml. It declares always_load files, axes, and references.on_demand.
- Very specific: line 13
Read every file listed under always_load.
≈ “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
Academic Experiments
将此 skill 视为"实验取证代理",目标是建立最短且可信的证据链,而不是尽量多跑实验。
Router Protocol
- Read
manifest.yaml. It declaresalways_loadfiles,axes, andreferences.on_demand. - Read every file listed under
always_load. These are the skill's binding rules — not reference material. - Apply the loaded material as constraints:
stance.mddefines non-negotiable rules, evidence type semantics, failure degradation, and scope.red-lines.mddefines absolute prohibitions. Do not negotiate these.output-contract.mddefines deliverables per mode and claim-readiness classification.anti-patterns.mddefines known failure modes and their correct alternatives.
- Detect the mode using the manifest's
modeaxis:experiment-evidence-pass,evidence-inventory-only, orminimal-reproducible-run. Align evidence type semantics to../shared/core/evidence-policy.md. - Echo the selected mode to the user before executing.
- Reach for
references/only when the manifest'sreferences.on_demandcondition is satisfied.
Modes
| Mode | Use when |
experiment-evidence-pass | Full audit: inventory + run + record + risk analysis |
evidence-inventory-only | Inventory existing artifacts only, no execution |
minimal-reproducible-run | Execute minimal reproducible command (e.g. eval existing checkpoint) |
Agent Dispatch
agents/experiment_agent.md is dispatched by academic-paper-writer orchestrator at Step 4. The agent may run experiments but must not modify project source code or data files, nor write paper prose independently.
Independent Use
| Input | Mode | Priority | Behavior |
repo_path + no run mode | experiment-evidence-pass | 2 (path trigger) | Full audit: inventory → env → minimal run → risk |
repo_path + "inspect only" | evidence-inventory-only | 1 (explicit) | Inventory only, no commands |
repo_path + specific command | minimal-reproducible-run | 1 (explicit) | Verify env → execute → record |
No repo_path | — | 3 (no input) | Ask path, or auto-detect entry files |
| Scenario | Recommended |
| Just auditing/reproducing evidence | This skill (standalone) |
| Writing results into paper prose | academic-paper-writer orchestrator |
| Draft results need verification | This skill → academic-reviser |
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.

