execute-plan

Executes structured implementation plans via orchestrated coder, verifier, and test-runner subagents.

Updated May 19, 2026
One-click install
npx skills add https://github.com/davidsunglee/pi-flow --skill execute-plan-davidsunglee
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: execute-plan
Source: https://github.com/davidsunglee/pi-flow/tree/main/packages/pi-flow-core/skills/execute-plan
Command: npx skills add https://github.com/davidsunglee/pi-flow --skill execute-plan-davidsunglee

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Executing a multi-task implementation plan by hand is error-prone: tasks have dependencies, tests regress silently, and ad-hoc verification is unreliable. This Skill orchestrates the full execution of a structured plan under docs/plans/, dispatching work to subagents in dependency waves with enforced verification and integration-test gates. ## Core Features & Use Cases - Wave-based parallel execution: Parses the plan's dependency graph, groups tasks into waves, and dispatches coder subagents in parallel with model-tier resolution and a hard parallelism cap. - Independent acceptance verification: Routes every completed task through a fresh verifier subagent that checks planner-authored acceptance criteria and Verify recipes, with a strict parser as the sole verdict classifier. - Baseline-aware regression gating: Captures a frozen baseline of pre-existing test failures, reconciles each wave's integration run against it, and offers debugger-first remediation for new failures. - Use Case: You have a generated plan file with 8 tasks across 3 waves. Run this Skill to set up a git worktree, execute each wave with TDD-enabled coder agents, verify acceptance criteria, commit per wave, and pass a final integration gate before branch completion. ## Quick Start Ask the assistant to execute the plan at docs/plans/my-feature.md using the execute-plan skill and confirm the proposed execution settings.

Frequently Asked Questions about execute-plan

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I execute a structured implementation plan with AI subagents?▼

Place a structured plan file under docs/plans/ and invoke the execute-plan skill. It validates the plan, groups tasks into dependency waves, dispatches coder subagents in parallel, verifies results with fresh verifier agents, and commits after each wave.

How does execute-plan verify that completed tasks meet acceptance criteria?▼

Each task is verified by a fresh verifier subagent that runs the plan's Verify recipes and inspects an orchestrator-assembled file set. A deterministic parser classifies the report as PASS, PASS_WITH_PROTOCOL_WARNINGS, or FAIL; the orchestrator never re-interprets results.

Does execute-plan require a git repository or worktree?▼

Yes, it stops immediately if the working directory is not a git repository. By default it creates a new git worktree with a branch derived from the plan filename, though it can reuse an existing worktree or feature branch.

What happens when integration tests fail after a wave?▼

Failures are reconciled against a frozen baseline captured before execution. New stable or non-reconcilable failures trigger a menu offering debugger-first remediation, continuing despite failures on intermediate waves, or stopping execution.

Why does execute-plan stop when a task is missing a Verify recipe?▼

A plan without complete Verify recipes is treated as a protocol error from generate-plan. The skill halts the wave and recommends regenerating the plan, since acceptance criteria cannot be verified deterministically without recipes.

What are the limitations of execute-plan's retry and remediation handling?▼

Blocked tasks support only three interventions: more context, a better model tier, or splitting into sub-tasks. Retries are budgeted per task, and refine-code remediation is capped at a configurable maximum of five iterations.