Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/write-plan
Use when a research design spec exists and the user is ready to translate it into a concrete implementation plan with phased tasks, artifacts, and verification criteria. Produces a research execution plan organized in canonical research phases — collection, preparation, analysis, robustness, writing, submission.
Passed 11 of 17 test cases with the skill, and 4 of 17 without it.
and 7 more in the full report
10 fixed4 passed either way7 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
When offering execution options to the user after writing the research plan, what two execution modalities must be offered?
| Checked | Without skill | With skill |
|---|---|---|
| The options present subagent-driven execution and inline execution. | Missed | Passed |
The instructions your model reads when the skill loads.
This skill writes a detailed, research-structured implementation plan from an approved design spec. It begins by invoking academic-baseline as the standing policy layer for the planning session. The plan is organized in research phases — not software phases — and every task has an expected artifact, verification criterion, and explicit skill routing. replication-driven-research is invoked as a hard constraint: every task must produce a versionable artifact and fit into an end-to-end reproducible pipeline.
docs/superpapers/specs/brainstorm to implementationdocs/superpapers/specs/YYYY-MM-DD-<topic>-design.md, written and approved via brainstormInvoke academic-baseline first. This resolves CLAUDE.superpapers.md via the walk-up Read (current working directory, then parent directories) and carries its settings and principles through the entire planning step.
Read the design spec in full. Understand every decision the brainstorm made — research question, identification strategy, data sources, models, robustness plan, submission target.
Decompose into tasks following the canonical research phases (see below). Aim for bite-sized tasks — 10 to 30 minutes of work each. If a task is a full day of work, split it.
For every task, specify all task-template fields (see template below). No placeholders. Every field filled explicitly.
Map dependencies. Later phases depend on earlier phases. Within a phase, tasks may be parallel or sequential — mark explicitly.
Assign skills deliberately. The Skills involved field is mandatory routing metadata, not decoration. Include academic-baseline on every task as the standing foundation, then add the domain skills actually needed. Any task involving target journal selection, author instructions, formatting, templates, blinding, cover letters, checklists, or submission portals must include journal-guidelines; if the outlet is not fixed yet, include journal-selection before journal-guidelines.
Apply replication-driven-research as a constraint. Every task must produce a versionable artifact. Every stochastic script must fix the seed. Every output must feed into an end-to-end pipeline.
Write the plan to docs/superpapers/plans/YYYY-MM-DD-<topic>-plan.md in English. The plan is a plugin artifact like the spec — English for consistency.
Self-review for coverage (every spec requirement has a task?), placeholders, name or type consistency across tasks, skill routing completeness, and scope fit (can this be one plan or does it need splitting?).
Offer execution options. Present execute-plan as the next step, with the choice between subagent-driven execution and inline execution.
test cases passed, vs 4 / 17 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.