一键导入
plan-harvest-and-retire
Retire a stalled or indefinitely-deferred plan without losing its value. Candidate scan, harvest, distribute, sweep, archive and codify.
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
Retire a stalled or indefinitely-deferred plan without losing its value. Candidate scan, harvest, distribute, sweep, archive and codify.
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。
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.
Automated pull request review for your repos. Five parallel agents, confidence scoring, convention-file compliance, and GitHub comment posting.
Use when the operator wants parallel sub-agent research aggregated into a structured report. Spawns up to 4 independent workers each investigating a different angle, then synthesizes findings for operator review. Research feeds decisions, does not trigger execution.
基于 SOC 职业分类
| name | plan-harvest-and-retire |
| description | Retire a stalled or indefinitely-deferred plan without losing its value. Candidate scan, harvest, distribute, sweep, archive and codify. |
| status | active |
| version | 1 |
Purpose: Retire a stalled or indefinitely-deferred Pending Plan without losing its value. The plan dies as an active queue item. Its value lives on in a durable home. Trigger: Operator-invoked. Also fires when a candidate surfaces during a regular planning review pass. Inputs: Target plan, or no target (run the candidate scan to find one). Per-plan archival consent before any move. Current durable homes: a deferred-items tracker, other active plans, your docs. Outputs: A read-only candidate list (scan mode), or a retired plan: value distributed, references repointed, file archived, any meta-lesson codified. Status: active
This skill assumes the Pending Plan object type already established by Pending Plan Implementation: a markdown note with frontmatter fields for status, scope and an activity log, tracked in a plans index. It also assumes a durable "deferred items tracker" as a landing zone for value that isn't ready to build yet, distinct from the active plan queue.
If you don't have a deferred-items tracker yet, the simplest version is a single markdown file with one entry per deferred idea: source, why it's deferred, the signal that would promote it back to active, and any context notes. Build that first. This skill needs somewhere for durable vision to land before it can retire a plan.
If your planning system lives in Linear, Jira or GitHub Issues, "retire" maps to closing the issue with a deferred label and copying the durable-vision pile into a linked tracker issue or wiki page rather than a markdown file. The read-scan-sort-distribute-sweep-archive sequence transfers regardless of where plans live.
A plan retires for value-plus-deferral, not staleness alone. A stale plan that still wants to be built is parked, not orphaned. A plan whose value is fully preservable elsewhere and whose build is gated on a far-off or undefined event is the retire candidate. The job is to move the value to a home where it survives and let the queue item go. Nothing of worth is lost. The queue gets lighter.
Lean by design. This is a periodic light sweep, not a backlog tool. Expect thin yield and keep the read cheap.
Data source. Read your plans index first, if you maintain one. Filter there before reading any plan body. Then grep only the shortlist bodies for deferral language. No new tooling required to run this skill.
Scan sequence:
proposed (or your equivalent early-stage status) and the last-modified date is roughly 30 or more days stale.Retire signal (needs the value-plus-deferral combination, not staleness alone):
proposed (or equivalent) and untouched for roughly 30 daysAnti-signal (not a candidate, leave it parked):
Output. A read-only candidate list: plan, signals matched, suggested durable home. No archival without per-plan consent. Present and stop. The operator selects which candidates proceed to Step 2.
For a confirmed candidate, read the full body once and sort everything of value into three piles:
This is judgment work. Read the plan for what it actually carries, not what its title implies.
#decision quick entry tag from Pending Plan Implementation, if you're using that convention.Every pile lands somewhere before the plan archives. A pile with no home is a blocker. Surface it rather than dropping it.
Run a full grep across your docs for the retiring plan's path and title. Every active reference resolves to the durable home, not the archive path. History stays as history.
retired, set a superseded-by field pointing to the durable home (tracker entry, successor plan or doc), set the implementation state to retired.Per-plan consent is required before this step runs. One plan at a time.
When the retire surfaces a rule worth keeping, land it canonically. Not every retire produces a rule. Most do not. When one does, the canonical home is a doc, a protocol or your team's conventions file, never a passing comment. Codifying a standing rule requires the same sign-off your team already uses for governance changes.
Operator-invoked by default. The candidate scan can surface as a one-line lens inside your weekly review or a periodic tooling-and-process check, whenever you want a light periodic surface rather than a standing cadence. The scan stays lean wherever it fires.
Default model per step. Set model explicitly on every subagent dispatch. Never silently inherit the orchestrator's model.
| Step | Default model | Rationale |
|---|---|---|
| 1. Candidate scan (index read, stale filter, deferral grep) | Haiku | Mechanical frontmatter and log read against a fixed signal set; no judgment |
| 1. Candidate confirmation (value-preservable, gate far-off) | Sonnet | Light judgment on whether a shortlisted plan truly qualifies |
| 2. Sort contents into three piles | Opus | Reads the plan for what it carries; judgment on value and placement |
| 3. Distribute to durable homes | Opus | Tracker-entry authoring, promotion-signal design, cross-plan placement |
| 4. Reference sweep and repoint | Haiku | Mechanical grep; Sonnet for ambiguous repoint calls |
| 5. Archive (file move, frontmatter edit, decision line) | Haiku | Deterministic file move and scoped frontmatter edits after consent |
| 6. Codify meta-lesson | Opus | Approval-gated governance prose; operator-facing sign-off |
Status is active. Execution is authorized. Operator invocation required. Do not self-invoke. The scan is read-only and may run freely. Any archival waits on per-plan consent.
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.
Pending Plan Implementation owns the plan object this skill retires. This skill is the closing bookend: Plan Challenger stress-tests a plan before it builds, Pending Plan Implementation executes it, and this skill retires it cleanly if it stalls instead of leaving it to rot in the queue.
Plan Challenger is the opening bookend. Challenge a plan before building it; harvest and retire it if the build never happens.
If Source Harvest isn't installed yet: Install Source Harvest via IGOS. If Pending Plan Implementation isn't installed yet: Install Pending Plan Implementation via IGOS. If Plan Challenger isn't installed yet: Install Plan Challenger via IGOS.
(Empty. Populated when execution mistakes occur during sessions.)