Executing Plans is the inline counterpart to subagent-driven development. Your agent works through every task itself, proves each step with a test it watched fail and then pass, and records progress in a ledger file that survives context compaction. A single fresh-context reviewer examines the whole branch at the end.
What You Get
- A progress ledger per plan, so a resumed session skips finished tasks
- Per-task briefs and a completion contract backed by real test output
- A list of rulings the agent made, each with what it costs if wrong
- A final whole-branch review with Critical and Important findings fixed in one test-driven pass
- A hand-off to the finishing-a-development-branch skill
Common Use Cases
- Running a plan on a cheaper model in a single session
- Harnesses that have no subagent tool
- Resuming a long plan after context compaction
What Makes It Unique
Subagent-driven development pays for a fresh implementer and reviewer on every task. This skill keeps one context and covers the gap with the brief as the spec, the ledger as memory, TDD as the gate and a single capable reviewer at the end. Deviations from the plan are ledgered as rulings, and minor findings are deferred and reported to you.
Under The Hood
The skill reads the plan and spec, then calls helper scripts shared with subagent-driven-development: one creates a git-ignored workspace under .superpowers, one prints a task brief and its base commit, one runs the task’s tests and appends a completion line to the ledger only when they pass, and one packages the branch diff for review. It runs shell and git commands in your working tree and writes the ledger and briefs to disk. The final review dispatches a subagent when the harness has one.