---
name: pm-task-hygiene-scanner
description: "Scan an Asana / ClickUp (or similar) project board and classify every open task — READY TO CLOSE / BLOCKED / STALE / DUPLICATE / ACTIVE / UNCLEAR — with confidence levels and evidence, plus a high-confidence \"propose to close\" batch. Permanently read-only: never closes, comments, reassigns, or moves anything. Use as a weekly/monthly hygiene pass or before a client review or handoff."
metadata:
  version: "1.1.0"
  scope: "standalone-template"
---

> Standalone draft/audit template. Not validated for OLLIE execution.


# PM Task Hygiene Scanner

## Who it's for

Any agency running client work through Asana, ClickUp, or a similar project-management tool — especially once a board has been live long enough to accumulate stale tasks nobody's quite sure are safe to close.

## What problem it solves

Every active client project ends up with a graveyard of tasks that are done-but-never-closed, blocked-on-someone-who-left, or duplicated across two different requests. Nobody wants to spend an afternoon triaging it, so it never happens — until the board is so noisy that real, urgent items get lost in it. This skill gives an agent the job of reading every open task and telling you, with evidence, what's actually safe to close.

## When to use it

- Weekly or monthly, as a standing hygiene pass on any active client project
- Before a client review or handoff, so the board reflects reality instead of six months of accumulated clutter
- Whenever a project starts to feel noisy and nobody trusts what's actually still open

## Required inputs

- Read-only access to the project's open tasks: title, description, assignee, last-updated date, and comment history
- A short description of what "done" typically looks like for this project's task types, if it's not obvious from the task titles alone

## Agent instructions

You are scanning an active project-management board for task hygiene. You
have READ-ONLY access. You may not close, comment on, reassign, edit, or
move a single task — this task produces a classification report only.

1. Pull every open task in the project: title, description, assignee,
   last-updated date, and comment history.
2. Classify each task into one of these categories, with your reasoning:
   - READY TO CLOSE — the work is clearly done based on comments/evidence,
     just never got closed out
   - BLOCKED — waiting on a specific named person or external dependency;
     name who or what it's waiting on
   - STALE / NO RECENT ACTIVITY — no update in a long time with no clear
     reason why, flagged for a human to check rather than assumed abandoned
   - DUPLICATE — clearly the same request as another open task; name both
   - ACTIVE / IN PROGRESS — genuinely being worked, leave alone
   - UNCLEAR — you don't have enough information to classify confidently
3. Attach a confidence level (high/medium/low) to every classification,
   especially READY TO CLOSE and DUPLICATE calls.
4. Produce an optional "propose to close" batch containing only the
   highest-confidence READY TO CLOSE items — never include medium- or
   low-confidence calls in this batch.
5. Do not close, comment on, reassign, or move a single task, at any
   confidence level. Produce the report and the proposed batch, and stop.
6. If you cannot confidently determine who owns a task or what its true
   status is, classify it as UNCLEAR rather than guessing.

## Expected output format

A classification table: Task | Category | Confidence | Evidence/Reasoning.

Followed by a separate, short "propose to close" list containing only the highest-confidence READY TO CLOSE items, clearly labeled as a proposal awaiting human approval.

## Human approval checkpoint

Nothing on the board changes as a result of this scan. A human reviews the proposed batch and closes items individually (or in bulk, at their own discretion) — the agent never closes, comments, reassigns, or moves anything itself, at any confidence level, ever.

## Safety rules

- Permanently read-only — this skill has no write access to the board at all, by design, not just by convention
- Every classification carries a confidence level and visible evidence
- The "propose to close" batch only ever contains high-confidence items — ambiguity always routes to UNCLEAR, never to a forced call

## What the agent must not do

- Close, comment on, reassign, or move any task
- Guess at task ownership when the assignee mapping is unclear — leave it UNCLEAR instead
- Treat a "looks done" impression as sufficient evidence without pointing to the specific comment or update that supports it

---

Part of the Outloop Agency Skill Pack ([outloop.co](https://outloop.co)). To turn this from a draft-only workflow into a safe runtime workflow across client workspaces, use Outloop to approve the right access, block wrong-client use, and keep secrets out of the agent context.