/execution-plan-manager

name: execution-plan-manager
description: 'Create, resume, analyze, or converge standard and deep frontend execution plans with traceable slices, evidence, and stop/resume state. Skip lightweight isolated edits.'
id: 'agents.skills.execution-plan-manager.skill'
title: 'Execution Plan Manager'
doc_type: 'skill'
layer: 'skill'
status: 'active'
publishable: true
local_only: false
skill: 'execution-plan-manager'
tags:
    - 'agents/skill-package'
    - 'agents/planning'
    - 'workflow/execution-plan'
parent:
    - '[[AGENTS|Canonical Agent Policy]]'
related:
    - '[[skills/goal-planner/SKILL|Goal Planner]]'
    - '[[skills/loop-workflow-planner/SKILL|Loop Workflow Planner]]'
    - '[[common/planning-rules|Planning Rules]]'
    - '[[common/planning-analysis-rules|Planning Analysis Rules]]'
    - '[[common/convergence-rules|Convergence Rules]]'
    - '[[common/execution-loops|Execution Loops]]'
    - '[[common/agent-loop-policy|Agent Loop Policy]]'
    - '[[common/token-budget-rules|Token Budget Rules]]'
    - '[[common/checkpoint-rules|Checkpoint Rules]]'
    - '[[templates/execution-plan|Execution Plan Template]]'
    - '[[templates/loop-workflow-contract|Loop Workflow Contract Template]]'
    - '[[common/prompt-intent-routing-rules|Prompt Intent Routing Rules]]'
    - '[[skills/project-context-adapter/SKILL|Project Context Adapter]]'
    - '[[skills/agent-rules-skill-author/SKILL|Agent Rules And Skill Author]]'
depends_on:
    - '[[AGENTS|Canonical Agent Policy]]'

Execution Plan Manager

Purpose

Create, resume, analyze, and converge execution plans for standard or deep frontend work after the goal is defined.

This skill prevents large tasks from becoming unbounded development sessions. It splits work into small verified slices, selects the minimum useful context budget, records stop/resume state when needed, and decides whether a slice needs a one-pass check, bounded retry loop, goal-based loop, independent review, or durable memory handoff.

When To Use

Use this skill only after `AGENTS.md` and `common/prompt-intent-routing-rules.md` classify the task as `Standard Workflow` or `Deep Workflow`.

Use this skill when:

• a goal contract exists or the user goal is already explicit;
• the task needs more than one implementation slice;
• the task may need stop/resume state;
• the task has dependencies, blockers, or ordered phases;
• the task should be split before coding;
• the user asks to continue an existing large task;
• the agent needs to compare completed work with the plan;
• the task may require measurable iteration and should be handed to `loop-workflow-planner`.

When Not To Use

Do not use this skill for `Lightweight Workflow` prompts, including one small bug, one obvious typo, one obvious type error, one small styling adjustment, one isolated component change, one direct file-scope edit, or one local refactor with clear boundaries.

Do not use this skill to implement code, install tools, scaffold projects, rewrite architecture, create tests, or perform visual QA.

If a lightweight task reveals hidden scope or repeated failure, escalate first using `common/prompt-intent-routing-rules.md`, then use this skill only if the escalated task is standard or deep.

Required Context

1. Read `AGENTS.md`. 2. Read `common/prompt-intent-routing-rules.md` when task scale is not obvious. 3. Read `common/planning-rules.md`. 4. Read `common/planning-analysis-rules.md` only in `analyze` mode. 5. Read `common/convergence-rules.md` only in `converge` mode. 6. Read `common/checkpoint-rules.md` when durable stop/resume state is needed. 7. Read `common/agent-loop-policy.md` when the plan may need measurable iteration. 8. Confirm the workflow level is `Standard Workflow` or `Deep Workflow`. 9. Read the current goal contract from the response, user request, or `project/active-goals.md` when present and relevant. 10. Read only project overlays needed to slice the task safely, such as `project/stack-profile.md`, `project/architecture-map.md`, `project/styling-profile.md`, or `project/verification-profile.md`. 11. Read affected source files only when slicing, analysis, or convergence cannot be done safely without them.

Tool Contract

• Use filesystem access only when a durable execution plan, progress log, loop contract, or stop/resume state must be written or updated. In `converge` mode, write only the append-only active-plan section allowed by `common/convergence-rules.md`.
• Do not use MCP installation checks from this skill.
• Do not use Browser, Playwright, Visual Diff, Figma, or design-tool MCP from this skill.
• Do not install packages or tools.
• Do not change application source files.
• Do not create testing workflows or testing files.

Workflow

1. Confirm task scale. If lightweight, stop and route back to the direct skill without creating an execution plan. 2. Confirm goal. Use an existing Goal Contract when available. If no goal exists and the task is ambiguous, route to `goal-planner` first. 3. Choose one explicit mode:

• `create`: make a new compact or durable plan;
• `resume`: continue the existing plan without changing stable identifiers;
• `analyze`: run the read-only pre-implementation checks in

`common/planning-analysis-rules.md`;

• `converge`: compare completed implementation and evidence with confirmed

intent under `common/convergence-rules.md`. 4. In `analyze` mode, inspect the confirmed goal, plan, decisions, scope boundaries, and minimum affected context. Return deterministic findings and stop the implementation handoff for `blocking` or `high` severity. Do not edit files or remediate findings. Then finish this skill. 5. In `converge` mode, first reconcile any changed current request in `create` or `resume` mode. Otherwise compare only active intent with implementation and verification evidence, return stable `F-###` findings, and never change implementation files. For a durable plan, append one convergence pass on