Menu

Individual skill

research-workflows

Execute the planning, code-review, comparison and export workflows behind the /plan, /review, /compare and /export commands as state machines over OpenScience session state, artifacts and compute. Internal support skill; reproduction, source audits and peer review are the reproduce, sources and peer-review core skills.

backend/cli/skills/research/research-workflows/SKILL.md

Read pinned source

Source commit a08e0ad0ad3d6448123039ee4659163eb2e35d99

What was checked

Matches the observed repository HEAD
Yes
Scanner rules are current
Yes
Repository scan complete
No
Skill scan complete
No
Observed behavior
Network or broad shell
Last checked
Nov 2, 2026

These are observations at the recorded commit, not a guarantee of safety. Review the source before running it.

Install this skill

Choose an agent in your Stack or copy a command below. Commands fetch the pinned commit and copy only this skill directory. Existing destinations are refused. Copying a command does not install anything.

An install command is unavailable: current rules, complete repository and skill scans, an eligible behavior level and a permitted license are required.

Repository context

Repository stars (not skill ratings)
4,352
Repository last updated
Nov 6, 2026
License
Apache-2.0

Observed agent integration: No specific integration evidence. Standard SKILL.md installation does not imply verified compatibility.

Files and excerpt

File metadata and permitted, redacted excerpts come from the recorded scan. Source links are pinned to that commit.

Browse files
Read excerpt

# Research Workflow Engine

Treat each workflow as a state machine, not a prose request. Follow the selected reference supplied with this skill. Do not substitute a generic answer for its admission checks, durable state updates, evidence collection, or terminal condition.

## State ownership

- Keep implementation steps in `planwrite` or `todowrite`; replace the complete list on every update and keep only genuinely active work in progress.
- Keep multi-stage scientific completion state in `research_contract`. Define the contract before expensive work, update a stage only at a real boundary, record every materially distinct failed candidate, and pass a check only after observing its evidence.
- Record reusable project method lessons only from a named evidence-backed trial. Treat a one-session lesson as tentative, reinforce it only from an independent session, and reject it explicitly when a later trial supplies counterevidence. Never turn a task-specific scientific claim into standing guidance.
- Keep files in the workspace or a writable connected project folder. Save durable Results through `artifact` with `action: save_file` only when the workflow calls for a reusable artifact, and keep exact claim lineage through provenance tools. Do not copy bulky file contents into conversational status text.
- Preserve a design manifest before execution when later claims depend on temporal freezing, and link producing runs to exact code, inputs, and outputs. A post-run checksum proves consistency, not preregistration.
- Treat jobs, kernels, reviewer findings, artifacts, and verification checks as authoritative session state. Never infer success from an assistant sentence when the recorded state disagrees.

## Execution protocol