Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Initialize a missing research Workspace from the repository template without overwriting existing Run state; use before Pipeline binding when the target under `workspaces/` has no valid core artifacts.
.claude/skills/willoscar-workspace-init/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-17 | ✗→✓ | ▲ Improved | -70% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 465% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -56% | 0% |
Initialization is a boundary: it creates Workspace-owned state and leaves existing state untouched unless overwrite is an explicit operator decision. It does not select a Workflow or execute a Unit.
workspaces/<name>/.STATUS.md, CHECKPOINTS.md, UNITS.csv, and DECISIONS.md.GOAL.md and queries.md.papers/, outline/, citations/, and output/ directories.Resolve the target path and reject the repository root. Inspect existing files before deciding whether initialization is missing, partial, or already complete. Never infer overwrite permission from a partial Workspace.
Completion criterion: the destination is a Workspace path and every existing file that could be replaced is known.
Copy the repository-owned template into the target. Default to create-missing; use --overwrite only after an explicit replacement decision.
Completion criterion: all missing base files and directories are materialized, and no pre-existing file was replaced without explicit permission.
Confirm that STATUS.md, UNITS.csv, CHECKPOINTS.md, and DECISIONS.md exist. Parse the UNITS.csv header and reject malformed CSV rather than leaving a superficially initialized Workspace.
Completion criterion: the four core files exist and UNITS.csv has the required schema.
Leave Workflow selection and Pipeline-specific Unit materialization to pipeline-router or the Pipeline adapter. The generic template is not evidence that a Goal has been routed.
Completion criterion: the Workspace can be passed to the router without requiring any state from chat history.
assets/workspace-template/ only when inspecting or changing the basetemplate contract.
pipelines/*.pipeline.md only after Workflow binding.research-pipeline-runner when the user asked to continue beyondinitialization.
bashuv run python .codex/skills/workspace-init/scripts/run.py \ --workspace workspaces/<name>
--workspace <path>: target Workspace.--overwrite: replace existing template-owned files after explicit approval.--unit-id, --inputs, --outputs, --checkpoint: Pipeline-runnercompatibility arguments.
Inspect the interface before use:
bashuv run python .codex/skills/workspace-init/scripts/run.py --help
Replace template files only after an explicit reset decision:
bashuv run python .codex/skills/workspace-init/scripts/run.py \ --workspace workspaces/<name> \ --overwrite
behavior, not an initialization failure.
workspaces/<name>/.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-17 | fail→pass | 20,694 | 2,099 | -90% | 1 | 1 | 0% | 3,311 | 1,009 | -70% | 0 | 0 | — |
case-01 | fail→fail | 6,676 | 4,868 | -27% | 1 | 1 | 0% | 965 | 954 | -1% | 0 | 0 | — |
case-02 | fail→fail | 4,625 | 6,150 | +33% | 1 | 1 | 0% | 535 | 959 | +79% | 0 | 0 | — |
case-03 | fail→fail | 19,034 | 4,866 | -74% | 1 | 1 | 0% | 3,101 | 895 | -71% | 0 | 0 | — |
case-04 | fail→pass | 7,049 | 4,088 | -42% | 1 | 1 | 0% | 224 | 1,265 | +465% | 0 | 0 | — |
case-05 | fail→pass | 5,906 | 3,433 | -42% | 1 | 1 | 0% | 814 | 1,329 | +63% | 0 | 0 | — |
case-06 | pass→pass | 13,729 | 3,235 | -76% | 1 | 1 | 0% | 1,917 | 1,203 | -37% | 0 | 0 | — |
case-07 | pass→pass | 11,493 | 5,061 | -56% | 1 | 1 | 0% | 1,801 | 1,534 | -15% | 0 | 0 | — |
case-08 | fail→pass | 25,265 | 2,686 | -89% | 1 | 1 | 0% | 992 | 1,107 | +12% | 0 | 0 | — |
case-09 | fail→pass | 11,721 | 2,134 | -82% | 1 | 1 | 0% | 2,230 | 992 | -56% | 0 | 0 | — |
case-10 | pass→pass | 6,620 | 3,863 | -42% | 1 | 1 | 0% | 957 | 1,267 | +32% | 0 | 0 | — |
case-11 | fail→pass | 7,179 | 3,377 | -53% | 1 | 1 | 0% | 1,001 | 1,241 | +24% | 0 | 0 | — |
case-12 | fail→pass | 11,638 | 3,867 | -67% | 1 | 1 | 0% | 1,971 | 1,365 | -31% | 0 | 0 | — |
case-13 | fail→pass | 11,968 | 1,975 | -83% | 1 | 1 | 0% | 1,921 | 991 | -48% | 0 | 0 | — |
case-14 | fail→pass | 13,530 | 1,491 | -89% | 1 | 1 | 0% | 1,961 | 875 | -55% | 0 | 0 | — |
case-15 | fail→pass | 9,972 | 1,533 | -85% | 1 | 1 | 0% | 1,727 | 936 | -46% | 0 | 0 | — |
case-16 | fail→pass | 12,257 | 3,527 | -71% | 1 | 1 | 0% | 1,853 | 1,291 | -30% | 0 | 0 | — |
case-18 | fail→pass | 9,653 | 2,680 | -72% | 1 | 1 | 0% | 1,334 | 1,098 | -18% | 0 | 0 | — |
case-19 | pass→pass | 7,310 | 2,936 | -60% | 1 | 1 | 0% | 1,015 | 1,142 | +13% | 0 | 0 | — |
case-20 | fail→pass | 7,426 | 3,121 | -58% | 1 | 1 | 0% | 1,001 | 1,164 | +16% | 0 | 0 | — |
case-21 | fail→pass | 10,126 | 5,382 | -47% | 1 | 1 | 0% | 1,634 | 1,518 | -7% | 0 | 0 | — |
case-22 | fail→pass | 12,946 | 5,709 | -56% | 1 | 1 | 0% | 2,114 | 1,640 | -22% | 0 | 0 | — |
case-23 | fail→pass | 8,020 | 2,032 | -75% | 1 | 1 | 0% | 1,214 | 984 | -19% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 18 counted toward the lift figure. The other 5 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +70 percentage points is the difference between those two pass rates over the 18 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.