shipyard-executing-plans

<!-- TOKEN BUDGET: 300 lines / ~900 tokens -->

Executing Plans

<activation>

When This Skill Activates

• You have a written implementation plan (from shipyard:shipyard-writing-plans or similar)
• Plan contains independent tasks suitable for agent dispatch or batch execution
• You need structured execution with two-stage review gates

Announce at start: "I'm using the executing-plans skill to implement this plan."

Natural Language Triggers

• "build this", "implement this", "build me this", "execute the plan", "run the plan"

</activation>

Overview

Execute implementation plans by dispatching fresh builder agents per task, with two-stage review after each: spec compliance review first, then code quality review. Can also run as batch execution with human checkpoints.

Core principle: Fresh agent per task + two-stage review (spec then quality) = high quality, fast iteration.

digraph when_to_use {
    "Have implementation plan?" [shape=diamond];
    "Tasks mostly independent?" [shape=diamond];
    "Stay in this session?" [shape=diamond];
    "Agent-driven execution" [shape=box];
    "Batch execution with checkpoints" [shape=box];
    "Manual execution or brainstorm first" [shape=box];

    "Have implementation plan?" -> "Tasks mostly independent?" [label="yes"];
    "Have implementation plan?" -> "Manual execution or brainstorm first" [label="no"];
    "Tasks mostly independent?" -> "Stay in this session?" [label="yes"];
    "Tasks mostly independent?" -> "Manual execution or brainstorm first" [label="no - tightly coupled"];
    "Stay in this session?" -> "Agent-driven execution" [label="yes"];
    "Stay in this session?" -> "Batch execution with checkpoints" [label="no - parallel session"];
}

<instructions>

The Process

Step 1: Load and Review Plan

1. Read plan file
2. Review critically - identify any questions or concerns about the plan
3. If concerns: Raise them with your human partner before starting
4. If no concerns: Create tasks via TaskCreate and proceed

Step 2: Execute Tasks

Agent-Driven Mode (preferred):

For each task, dispatch a fresh builder agent:

digraph process {
    rankdir=TB;

    subgraph cluster_per_task {
        label="Per Task";
        "Dispatch builder agent" [shape=box];
        "Builder asks questions?" [shape=diamond];
        "Answer questions, provide context" [shape=box];
        "Builder implements, tests, commits, self-reviews" [shape=box];
        "Dispatch spec reviewer agent" [shape=box];
        "Spec reviewer confirms code matches spec?" [shape=diamond];
        "Builder fixes spec gaps" [shape=box];
        "Dispatch code quality reviewer agent" [shape=box];
        "Code quality reviewer approves?" [shape=diamond];
        "Builder fixes quality issues" [shape=box];
        "Mark task complete" [shape=box];
    }

    "Read plan, extract all tasks, create via TaskCreate" [shape=box];
    "More tasks remain?" [shape=diamond];
    "Dispatch final reviewer for entire implementation" [shape=box];
    "Use shipyard:git-workflow to complete" [shape=box style=filled fillcolor=lightgreen];

    "Read plan, extract all tasks, create via TaskCreate" -> "Dispatch builder agent";
    "Dispatch builder agent" -> "Builder asks questions?";
    "Builder asks questions?" -> "Answer questions, provide context" [label="yes"];
    "Answer questions, provide context" -> "Dispatch builder agent";
    "Builder asks questions?" -> "Builder implements, tests, commits, self-reviews" [label="no"];
    "Builder implements, tests, commits, self-reviews" -> "Dispatch spec reviewer agent";
    "Dispatch spec reviewer agent" -> "Spec reviewer confirms code matches spec?";
    "Spec reviewer confirms code matches spec?" -> "Builder fixes spec gaps" [label="no"];
    "Builder fixes spec gaps" -> "Dispatch spec reviewer agent" [label="re-review"];
    "Spec reviewer confirms code matches spec?" -> "Dispatch code quality reviewer agent" [label="yes"];
    "Dispatch code quality reviewer agent" -> "Code quality reviewer approves?";
    "Code quality reviewer approves?" -> "Builder fixes quality issues" [label="no"];
    "Builder fixes quality issues" -> "Dispatch code quality reviewer agent" [label="re-review"];
    "Code quality reviewer approves?" -> "Mark task complete" [label="yes"];
    "Mark task complete" -> "Extract micro-lesson (best-effort)";
    "Extract micro-lesson (best-effort)" -> "More tasks remain?";
    "More tasks remain?" -> "Dispatch builder agent" [label="yes"];
    "More tasks remain?" -> "Dispatch final reviewer for entire implementation" [label="no"];
    "Dispatch final reviewer for entire implementation" -> "Dispatch auditor for security review";
    "Dispatch auditor for security review" -> "Critical security findings?" [shape=diamond];
    "Critical security findings?" -> "Builder fixes security issues" [label="yes"];
    "Builder fixes security issues" -> "Dispatch auditor for security review" [label="re-audit"];
    "Critical security findings?" -> "Dispatch simplifier for complexity review" [label="no / user defers"];
    "Dispatch simplifier for complexity review" -> "High priority simplifications?" [shape=diamond];
    "High priority simplifications?" -> "Builder implements simplifications" [label="user chooses fix"];
    "Builder implements simplifications" -> "Dispatch simplifier for complexity review" [label="re-check"];
    "High priority simplifications?" -> "Use shipyard:git-workflow to complete" [label="no / user defers"];
}

Micro-lesson (best-effort): After marking complete, extract one line: what surprised you, what failed first, or what you'd do differently. Append to .shipyard/phases/{N}/MICRO-LESSONS.md as - [PLAN-{W}.{P} Task {T}] <takeaway>. Pass the file contents to the next builder agent as context. If no clear takeaway, skip silently. Metrics Collection (best-effort): Also check agent output for <!-- context: ... -->. If found, append to .shipyard/phases/{N}/AGENT-METRICS.md: {ISO timestamp} | {agent-type} | {task-label} | turns={N} | compressed={yes|no} | complete={yes|no}. Skip silently if missing.

Batch Mode (separate session):

Default: First 3 tasks per batch.

For each task:

1. Mark as in_progress
2. Follow each step exactly (plan has bite-sized steps)
3. Run verifications as specified
4. Mark as completed

When batch complete:

• Show what was implemented
• Show verification output
• Say: "Ready for feedback."

Based on feedback: apply changes if needed, execute next batch, repeat until complete.

Two-Stage Review Pattern

Stage 1: Spec Compliance Review

• Does the code match the plan's specification?
• Are all requirements met?
• Is there anything extra that wasn't requested?
• Are verification criteria satisfied?

Stage 2: Code Quality Review

• Is the code well-structured?
• Are there any bugs or edge cases missed?
• Is naming clear and consistent?
• Are tests comprehensive?

IMPORTANT: Always complete spec compliance before code quality. Wrong order wastes time reviewing quality of code that doesn't meet spec.

Step 3: Post-Completion Quality Gates

After the final reviewer approves the entire implementation, run these quality gates:

Security Audit

Dispatch an auditor agent (subagent_type: "shipyard:auditor") with:

• Git diff of all files changed during plan execution
• All task summaries and context
• Dependency manifests if any dependencies were added/changed
• Working directory, current branch, and worktree status
• Follow Model Routing Protocol — resolve model from model_routing.security_audit (default: sonnet). See docs/PROTOCOLS.md for details.

If CRITICAL findings exist:

1. Display the critical findings to the user
2. User decides: fix now (dispatch builder with audit feedback) / defer (append to ISSUES.md) / acknowledge and proceed
3. If fixing, re-run audit after fixes

Simplification Review

After the audit, dispatch a **simplifi