Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →notque/subagent-driven-development
Fresh-subagent-per-task execution with two-stage review gates.
Passed 12 of 16 test cases with the skill, and 8 of 16 without it.
and 4 more in the full report
7 fixed9 passed either way4 still failing2 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
We extracted 5 tasks from `plans/billing-engine.md`. What initial status marker should be applied to each task in the TodoWrite tracker before execution starts?
| Checked | Without skill | With skill |
|---|---|---|
| Marks all initial tasks as pending in the todo tracker | Missed | Passed |
The instructions your model reads when the skill loads.
Execute a plan with fresh implementers. Planning owns the plan format; this skill owns task dispatch and integration. Check requirements before code quality. Use the review lane in skills/process/pr-workflow/SKILL.md to choose direct or independent review; do not add another roster to a completed review.
Step 1: Read plan and extract tasks
Read the plan once. For each task, retain its full text, files, dependencies, and verification commands. Pass these inline rather than sending workers back to the plan.
Step 2: Create TodoWrite
Track all tasks as pending, in progress, or complete with TodoWrite or the harness's task tracker. Reuse the existing plan; do not create a second status artifact.
Step 3: Gather scene-setting context
Capture git status, project conventions from CLAUDE.md, relevant code patterns, dependencies, and setup requirements. Capture BASE_SHA with git rev-parse HEAD before the first implementer: it anchors the final integration diff. Include the context each worker needs in its dispatch.
Step 4: Determine parallel vs sequential dispatch (scope-overlap check)
python3 scripts/check-scope-overlap.py \
--tasks '[{"id":"task-1","scope":["path/a.py"]},{"id":"task-2","scope":["path/b.py"]}]' \
--human # --check: exit 0 for no conflicts, 1 for conflicts
Each task has an id and scope path list. Directories include descendants; "readonly": true marks read-only tasks. Use the reported parallel groups, then account for dependencies: serialize overlapping writes and tasks that consume earlier results. Independent groups may run in parallel.
Gate: Full tasks, BASE_SHA, context, dependencies, and checked file scopes are ready.
Step 1: Mark task in_progress
Update the task tracker.
Step 2: Dispatch implementer subagent
test cases passed, vs 8 / 16 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.