# Research Experiment > >- - **Type:** Skill - **Install:** `agentstack add skill-v0lka-skills-research-experiment` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [v0lka](https://agentstack.voostack.com/s/v0lka) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [v0lka](https://github.com/v0lka) - **Source:** https://github.com/v0lka/skills/tree/main/engineering/research-experiment ## Install ```sh agentstack add skill-v0lka-skills-research-experiment ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Experiment Execution and Recording Guide the design, execution tracking, and result recording of experiments that test research hypotheses. ## Context Each iteration in the methodology tests one hypothesis through an experiment. The experiment must produce a clear signal: the hypothesis is confirmed, refuted, or the result is inconclusive. The emphasis is on minimality — the goal is to get an answer, not to build a product. ## Experiment Design Principles 1. **Minimality**: create the smallest prototype or analysis that can confirm or refute the hypothesis. Resist scope creep. 2. **Reproducibility**: key experiments (those that feed into the final report) must be reproducible by other engineers without the author's involvement. This means: documented environment, launch instructions, and setup scripts. 3. **Timebox awareness**: the experiment runs within the hypothesis timebox. If the timebox is about to expire and the result is unclear, that itself is a result — do not extend without a deliberate decision. ## Procedure ### Step 1 — Read the hypothesis card Locate and read the hypothesis card (`hypotheses/H-NNN.md`). Extract: - The statement being tested. - The verification criterion (what counts as confirmation). - The timebox (how long is allocated). - The current status (must be `open` or `in-progress`). If the status is not `open` or `in-progress`, stop and inform the researcher that this hypothesis is not in an experimentable state. ### Step 2 — Plan the experiment Help the researcher design a minimal experiment: 1. **Approach**: what will be built, analyzed, or measured? Suggest the simplest viable approach. 2. **Environment**: what OS, language versions, tools, and dependencies are needed? Capture these now — they form the reproducibility requirements. 3. **Expected output**: what artifact will demonstrate the result (PoC, benchmark data, log output, counterexample)? 4. **Completion signals**: specifically, what outcome confirms the hypothesis and what outcome refutes it? Map the verification criterion to concrete observables. If the verification criterion in the card is ambiguous, propose a more precise operationalization and update the card with the researcher's approval. ### Step 3 — Set up the environment Assist with environment setup: - Suggest Dockerfile, Makefile, shell scripts, or other setup automation as appropriate for the experiment's complexity. - For trivial experiments (single script, no unusual dependencies), a note in the card's Experiment Notes section is sufficient. - For complex setups, recommend creating a dedicated directory in the project's prototype/tool area and versioning it. Document the environment in the hypothesis card's Experiment Notes section. ### Step 4 — Track progress During the experiment, help the researcher track progress: - Periodically check timebox status. If more than 75% of the timebox has elapsed without a clear signal, alert the researcher and suggest: - Simplifying the experiment scope. - Identifying what is blocking progress. - Considering whether the timebox should trigger a `kill` or `pivot` decision. - Record intermediate findings in the Experiment Notes section of the hypothesis card. These notes are free-form but should capture key observations, dead ends, and unexpected discoveries. ### Step 5 — Record the result When the experiment concludes (by result or by timebox expiry): 1. **Determine the outcome**: - **Confirmed** — the verification criterion is met. - **Refuted** — the verification criterion is demonstrably not met. - **Inconclusive** — neither confirmed nor refuted within the timebox. Document what would be needed for a definitive result. 2. **Update the hypothesis card** via the `research-hypothesis` skill: - Set Status to the appropriate value. - Fill the Result section with: findings, link to prototype/proof, artifact list. - Set the Completed date. 3. **Document reproducibility** (for experiments that will enter the final report): - Environment description (OS, language versions, dependencies). - Launch instructions (step-by-step). - Setup scripts or configuration files. ### Step 6 — Hand off to decision After recording the result, inform the researcher that the iteration is ready for the decision step. Suggest invoking the `research-decision` skill to determine the next action (continue / pivot / kill / fork). ## Reproducibility Checklist For experiments that are candidates for the final report, verify: - [ ] Environment versions documented (OS, runtime, key libraries). - [ ] Dependencies listed (package file, requirements.txt, go.mod, etc.). - [ ] Launch instructions written and tested. - [ ] Setup automation exists if the environment is non-trivial. - [ ] Output artifacts are saved and linked from the hypothesis card. Intermediate experiments that were discarded do not need to meet this bar. ## Relation to Other Skills - Follows `research-hypothesis` (a hypothesis card must exist before an experiment can begin). - Results feed into `research-decision` for next-action determination. - Reproducibility artifacts are inventoried by `research-synthesis`. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [v0lka](https://github.com/v0lka) - **Source:** [v0lka/skills](https://github.com/v0lka/skills) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-v0lka-skills-research-experiment - Seller: https://agentstack.voostack.com/s/v0lka - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.