SkillValidated
Plan-then-execute — a working procedure for non-trivial tasks
A skill your agent invokes when a task is non-trivial: write the spec and acceptance check, produce a step-ordered plan, get it approved, then execute one step at a time with a checkpoint after each. The procedure form of the plan-first rule.
$cosq add plan-then-executeFree, works signed out. No CLI? Copy the file below into your repo — same result.
Why this one
Experienced developers plan before they generate (UChicago); spec → plan → implement → verify is the reliability consensus for agent output.
How to use it
- 1Invoke on any task that touches more than one file or one behavior
- 2Approve the plan before the first edit
What gets installed
Pick your tool to preview the exact file cosq add writes — byte-for-byte. Prefer copy-paste? Use the copy button on the file.
.claude/skills/plan-then-execute/SKILL.md(created)--- name: plan-then-execute description: Use when starting any non-trivial coding task: produce a spec and a step-ordered plan, get approval, then execute step by step with checkpoints. --- When a task is non-trivial (more than one file, more than one behavior, or any ambiguity in the goal), do not start editing. Instead: 1. **Spec** — restate the goal and the acceptance check in one or two sentences. If the acceptance check is unclear, ask before proceeding. 2. **Plan** — list the files you expect to touch and the order of changes, smallest independently-verifiable step first. 3. **Checkpoint** — present the spec and plan, and wait for approval. 4. **Execute** — one step at a time. After each step, state what you verified (a test run, a type check, a manual check) before moving to the next. 5. **Close** — confirm the acceptance check from step 1 explicitly. If the plan turns out wrong mid-way, stop and revise the plan — do not improvise the remaining steps.