一键导入
pending-plan-implementation
Execute an approved plan (or approved subset) using the plan note as the implementation context, breadcrumb record and outcome summary.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Execute an approved plan (or approved subset) using the plan note as the implementation context, breadcrumb record and outcome summary.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when a long-form manuscript (book chapter, ebook, multi-chapter playbook, long-form digital product) needs an editing pass to identify and remove AI writing tells. Sweeps across 8 pattern categories, assembles a structured edit packet for operator review, and applies approved edits.
Use when you have spare capacity and want to set your system improving without active attention. Operator-invoked only, never scheduled. Safe, additive, reversible hygiene work runs under a threshold model with a permanent floor of actions the agent never takes on its own.
Use when archiving several workspace artifacts at once with per-artifact approval gates before any archive move executes.
Use when deciding whether a piece of deterministic, no-judgment work should route to a secondary AI provider instead of your primary interface. Covers the provider catalog, the five-condition eligibility test and the dispatch script.
Retire a stalled or indefinitely-deferred plan without losing its value. Candidate scan, harvest, distribute, sweep, archive and codify.
Automated pull request review for your repos. Five parallel agents, confidence scoring, convention-file compliance, and GitHub comment posting.
| name | pending-plan-implementation |
| description | Execute an approved plan (or approved subset) using the plan note as the implementation context, breadcrumb record and outcome summary. |
| status | active |
| version | 1.1 |
Purpose: Execute an approved plan (or approved subset) using the plan note as the implementation context, breadcrumb record and outcome summary. Trigger: User selects a plan to implement and grants execution approval for all or part of the plan. Inputs: Target plan link, approved execution scope, approval boundaries, dependencies/links, desired outcome for this run. Outputs: Implemented changes (if approved), updated plan note (snapshot + execution log + outcome) and a proposal for status/archival updates if applicable. Status: active
This skill is a starting point, not a finished tool. It assumes a specific plan artifact: a markdown note with frontmatter fields for status, scope and an activity log. Your planning system may live somewhere entirely different.
The core discipline transfers to any format. Read the plan before executing. Confirm scope before touching anything. Log what you did as you do it. Surface proposed status changes for approval rather than writing them unilaterally. That sequence is the bone. The artifact format is the muscle, and muscle is yours to shape.
If your plans live in Linear, Jira or GitHub Issues. Replace "read the plan note" in Step 1 with a fetch of the issue or epic. Replace "update the plan note" in Steps 2 and 5 with comments on the issue and field updates on the record. The scope-confirm gate (Step 3) and the execution log (Step 4) translate directly.
If your plans live in Notion. The frontmatter fields become database properties. Status advances via property update rather than frontmatter edit. The activity log becomes a Notion comment thread or a child page. The skill's step sequence holds.
If your plans live in plain markdown without frontmatter. Drop the frontmatter requirement. Use a ## Status heading section instead. The read-confirm-execute-log-propose sequence doesn't depend on frontmatter. It depends on reading intent before acting and writing outcomes after.
Status vocabulary. The reference status ladder (proposed, approved, ready-for-execution, implemented) is one convention. Your system may use different terms. Status advances in one direction. "Not yet approved" blocks execution. The final state captures a breadcrumb with a timestamp and what shipped.
Edit the Steps to name your planning artifact format before first use. The skill is a substrate. You are the operator. Make it yours.
Run these gates before Step 1. All must pass before execution begins.
proposed or draft, stop. Execution is not authorized. Surface the status to the operator and ask what approval is needed before proceeding.Confirm the target plan and read back plan intent, risk, affected areas and current breadcrumb context.
Surface the plan's stated goal and its "Distinct from" contrast if one is present. "Building toward X. Distinct from Y." This is the within-session drift detection point. If the plan does not have an explicit contrast, ask the operator to name one before execution begins.
Document Implementation Snapshot (When Executing) and Approved Execution Scope (This Run).
Confirm exactly what is approved to implement now and what remains out of scope.
Execute approved work and log actions in Implementation Actions (Execution Log).
Skill Handover Map. Before executing, scan the plan for a ## Skill Handover Map section. If present, treat it as canonical phase-by-phase orchestration. For each named phase, invoke the specified skill with the context payload the map provides, and capture output before advancing to the next phase. Plans without a Skill Handover Map execute normally through the steps below.
Archival discipline. When the plan includes a file archival step, never write a new copy and delete the original as two separate operations. Move or rename directly. rm commands are for actual deletions only, not for clearing a replaced copy.
Update Evidence / Implementation Refs and Implementation Outcome (This Run).
Session Boundary Block. For multi-session plans, write a structured re-entry anchor at the end of each session's log entry:
### Session Boundary: YYYY-MM-DD
**Completed this run:** [one sentence]
**Decision point:** [any approval or gate holding next session]
**Next move:** [first action next session picks up]
**Live constraints:** [blockers, dependencies, anything that changed]
This block is what the Pre-Execution Checklist Step 3 reads at next-session entry. Keep it current.
Propose implementation_state and any status/archival changes (approval-gated).
Audit gate. Plans spanning 3 or more phases get a Phase Audit as the default exit gate before archival. Run two concurrent passes: a checklist pass (does the output match the plan's stated goal and scope?) and a challenger pass (does any finding contradict the plan's assumptions?). Smaller plans get an optional offer in one line ("Audit warranted?"). Do not offer for operational outputs (scripts, logs, sync results, backup records).
## Refinements in the audited document. Keep it to the durable insight, not a transcript of the full report. If no, close and move on.Read back the implementation summary and apply approval-gated changes only with explicit consent.
#field#ai#learning#decisionUse this block for plans that involve building or modifying software. Skip it for documentation, governance and content plans.
Phase 0: Documentation Discovery. Before planning any implementation that touches an external API, library or tool, find the actual current API reference. Do not plan against assumed method signatures or remembered behavior. A single documentation pass at the start prevents cascading rework.
Subagent Reporting Contract. When dispatching sub-agents for research or implementation phases, require each to return: sources consulted, concrete findings (not conclusions), copy-ready code snippets where relevant, and a confidence note. Reject and re-prompt any sub-agent that returns conclusions without cited sources.
Per-Phase Verification Before Commit. Before committing each phase's output, grep for the anti-patterns identified in Phase 0 documentation discovery. Do not advance phases on unverified work. Verification is part of the phase, not an optional close step.
Source Harvest is the gateway skill for systematic pattern extraction from external repos and tools. Many users adopt Source Harvest first, then layer additional skills like this one on top.
Plan Challenger runs an adversarial pre-build pass on the plan before this skill executes. Strong pairing for high-stakes implementations.
If Source Harvest isn't installed yet: Install Source Harvest via IGOS. If Plan Challenger isn't installed yet: Install Plan Challenger via IGOS.
Dispatch the cheapest model that does the job well. Before each delegated step, ask whether a smaller model would produce equivalent output.
| Work type | Model |
|---|---|
| Mechanical lookups, deterministic commands, structured extraction against a spec | Haiku |
| Multi-step synthesis, drafting, diagnosis, most worker dispatch | Sonnet |
| Architectural judgment, plan design, judgment-dense synthesis | Opus |
Per-step defaults: Pre-Execution Checklist (mechanical field checks) routes to Haiku. Steps 1 and 3 (operator-facing scope reads and confirmations) route to Opus. Steps 2, 4 and 5 (execution and logging) route to Sonnet. Step 7 audit passes route to Sonnet for checklist and Opus for challenger.
Set the model explicitly on every subagent dispatch. Never silently inherit the top tier.
(Empty. Populated when execution mistakes occur during sessions.)