Skills · trackers

`backlog-triage`

Triage a whole backlog export from Jira, Linear or GitHub Projects (CSV, JSON or an MCP tool) — duplicate and near-duplicate pairs with the reason, items missing acceptance criteria or an owner, stale items, oversized items to split, priority conflicts, and a proposed ordering with a reason per item — ending in a change list the user can apply. A stdlib script computes the mechanical checks. Use when the user wants to clean up, groom or refine a backlog, find duplicate tickets, or says "triage our backlog", "what should we pull next?".

Pack trackersTier standardJob Build and review softwarev0.1.3

npx skilldrop-cli install backlog-triage

Other ways to install

With the sibling skills it hands off to:

npx skilldrop-cli install backlog-triage --with-related

The whole pack it belongs to:

npx skilldrop-cli install --pack trackers

By hand, from a clone of the repo:

cp -R packs/trackers/skills/backlog-triage ~/.claude/skills/

In Claude Code, with the trackers plugin installed, invoke it as:

/trackers:backlog-triage

Try it

A realistic prompt from the skill's acceptance evals:

Here's the Northwind checkout backlog exported from Jira: examples/northwind-backlog.csv. Refinement is Thursday. Find the duplicates, tell me what isn't ready, and give me an order for what we pull next. Anything untouched for 30 days counts as stale.

What a good result looks like

Numbers come from the data. Counts, ages and estimates are the script's output or a stated count from the export, never an estimate.
Every duplicate names the pair and why. Which two keys, what makes them the same work, and which one survives.
Readiness is explicit. Each item in the ordering says whether it can be pulled today, and if not, what's missing.
Each position in the ordering has a reason tied to the ordering method, not "feels important".
Priority changes are proposals, each with the facts behind it, addressed to whoever owns priority.
The change list can be applied as written: key, change, reason, one row per action.
Nothing is invented. No owner, date, estimate or customer count that isn't in the export or the user's message.

Example output

Example: triaging the Northwind checkout backlog · the skill's own worked example, source

Worked example

The user exported the Northwind checkout team's Jira backlog to CSV and asked: "Triage this before Thursday's refinement. Who's missing, what's a duplicate, what should we pull next?" The export is northwind-backlog.csv: 20 issues, Jira's default CSV columns, dates in Jira's 02/Jun/26 9:14 AM format.

Step 1: run the script

python3 scripts/triage_backlog.py examples/northwind-backlog.csv --as-of 2026-10-01 \
  -o triage.md --changes changes.csv

Step 2: the script's output, unedited

triage.md:

# Backlog triage findings: northwind-backlog.csv

As of 2026-10-01 · 20 items in export · 18 open · 2 done or closed (skipped)
Thresholds: stale > 30 days · similarity >= 0.75 · oversized estimate > 8

Columns used: key='Issue key', title='Summary', description='Description', status='Status', owner='Assignee', priority='Priority', estimate='Custom field (Story Points)', created='Created', updated='Updated', labels='Labels'

| Check | Count |
|---|---:|
| Duplicate candidate pairs | 3 |
| No owner | 8 |
| No acceptance criteria | 11 |
| Stale (> 30 days) | 10 |
| Oversized | 3 |
| Priority conflicts | 11 |

## Duplicate candidates

| Pair | Score | Titles | Shared words |
|---|---:|---|---|
| CHK-116 / CHK-117 | 0.87 | Persist cart across devices for signed-in users / Persist the cart across devices for logged-in shoppers | across cart devices persist |
| CHK-103 / CHK-109 | 0.86 | Apple Pay button missing on Safari 17 / Apple Pay button not showing in Safari | apple button pay safari |
| CHK-101 / CHK-104 | 0.80 | Guest checkout without creating an account / Allow guest checkout without an account | account checkout guest without |

Title similarity only. Read both descriptions before merging.

## No owner

| Key | Title | Status | Priority | Last update |
|---|---|---|---|---|
| CHK-101 | Guest checkout without creating an account | To Do | Highest | 2026-06-10 |
| CHK-104 | Allow guest checkout without an account | Backlog | Medium | 2026-07-19 |
| CHK-109 | Apple Pay button not showing in Safari | Backlog | Medium | 2026-08-26 |
| CHK-111 | Split shipment tracking emails | Backlog | Medium | 2026-02-15 |
| CHK-114 | Klarna pay-later option | Backlog | Highest | 2026-08-22 |
| CHK-115 | Gift message at checkout | Backlog | Lowest | 2025-11-11 |
| CHK-117 | Persist the cart across devices for logged-in shoppers | Backlog | Low | 2026-06-18 |
| CHK-119 | Accessibility fixes on the payment form | To Do | High | 2026-09-21 |

## No acceptance criteria

| Key | Title | Status | Priority | Owner |
|---|---|---|---|---|
| CHK-101 | Guest checkout without creating an account | To Do | Highest | - |
| CHK-104 | Allow guest checkout without an account | Backlog | Medium | - |
| CHK-106 | Rebuild the payments service on the new platform | Backlog | High | Tom Okafor |
| CHK-109 | Apple Pay button not showing in Safari | Backlog | Medium | - |
| CHK-110 | Address autocomplete for UK postcodes | To Do | Low | Marco Rossi |
| CHK-111 | Split shipment tracking emails | Backlog | Medium | - |
| CHK-114 | Klarna pay-later option | Backlog | Highest | - |
| CHK-115 | Gift message at checkout | Backlog | Lowest | - |
| CHK-117 | Persist the cart across devices for logged-in shoppers | Backlog | Low | - |
| CHK-119 | Accessibility fixes on the payment form | To Do | High | - |
| CHK-120 | Checkout analytics dashboard | Backlog | Highest | Marco Rossi |

## Stale (no update in more than 30 days)

| Key | Title | Status | Priority | Days since update |
|---|---|---|---|---|
| CHK-115 | Gift message at checkout | Backlog | Lowest | 324 |
| CHK-111 | Split shipment tracking emails | Backlog | Medium | 228 |
| CHK-106 | Rebuild the payments service on the new platform | Backlog | High | 182 |
| CHK-110 | Address autocomplete for UK postcodes | To Do | Low | 154 |
| CHK-120 | Checkout analytics dashboard | Backlog | Highest | 142 |
| CHK-101 | Guest checkout without creating an account | To Do | Highest | 113 |
| CHK-117 | Persist the cart across devices for logged-in shoppers | Backlog | Low | 105 |
| CHK-104 | Allow guest checkout without an account | Backlog | Medium | 74 |
| CHK-114 | Klarna pay-later option | Backlog | Highest | 40 |
| CHK-109 | Apple Pay button not showing in Safari | Backlog | Medium | 36 |

## Oversized (split before scheduling)

| Key | Title | Status | Priority | Why |
|---|---|---|---|---|
| CHK-101 | Guest checkout without creating an account | To Do | Highest | estimate 13 > 8 |
| CHK-106 | Rebuild the payments service on the new platform | Backlog | High | estimate 21 > 8 |
| CHK-114 | Klarna pay-later option | Backlog | Highest | estimate 13 > 8 |

## Priority conflicts

- Priority inflation: 5 of 18 prioritised open items (28%) are at the top priority.
- CHK-101 (Guest checkout without creating an account): priority Highest but no owner
- CHK-101 (Guest checkout without creating an account): priority Highest but no update in 113 days
- CHK-106 (Rebuild the payments service on the new platform): priority High but no update in 182 days
- CHK-114 (Klarna pay-later option): priority Highest but no owner
- CHK-114 (Klarna pay-later option): priority Highest but no update in 40 days
- CHK-119 (Accessibility fixes on the payment form): priority High but no owner
- CHK-120 (Checkout analytics dashboard): priority Highest but no update in 142 days
- CHK-116 (Persist cart across devices for signed-in users): likely duplicate of CHK-117 but priority Medium vs Low
- CHK-103 (Apple Pay button missing on Safari 17): likely duplicate of CHK-109 but priority High vs Medium
- CHK-101 (Guest checkout without creating an account): likely duplicate of CHK-104 but priority Highest vs Medium

## Priority distribution (open items)

| Priority | Open items |
|---|---:|
| Highest | 5 |
| High | 5 |
| Medium | 5 |
| Low | 2 |
| Lowest | 1 |

changes.csv (first 6 of 35 rows):

key,action,field,proposed_value,reason
CHK-117,confirm-duplicate,status,Duplicate of CHK-116,title similarity 0.87 with CHK-116
CHK-109,confirm-duplicate,status,Duplicate of CHK-103,title similarity 0.86 with CHK-103
CHK-104,confirm-duplicate,status,Duplicate of CHK-101,title similarity 0.80 with CHK-101
CHK-101,assign-owner,assignee,,open item with no owner
CHK-104,assign-owner,assignee,,open item with no owner
CHK-109,assign-owner,assignee,,open item with no owner

Step 3: the triage, after reading the descriptions

Every count below comes from the script output above. The duplicate verdicts, the priority proposals and the ordering are judgment, and each one gives its reason.

# Backlog triage: Northwind checkout, as of 2026-10-01

18 open items. 3 duplicate pairs, 8 with no owner, 11 with no acceptance criteria,
10 stale (over 30 days), 3 oversized. 5 of 18 open items (28%) are at Highest.

## Duplicates (confirmed by reading both descriptions)

| Keep | Close | Why they're the same | Why this one survives |
|---|---|---|---|
| CHK-116 | CHK-117 | Both ask for the cart to follow a signed-in shopper across devices | CHK-116 has an owner, acceptance criteria and an update 7 days ago; CHK-117 has none of these |
| CHK-103 | CHK-109 | Same symptom: the Apple Pay button doesn't show in Safari | CHK-103 names the version (Safari 17), the report count (14) and acceptance criteria. Copy CHK-109's customer note into CHK-103 as a comment |
| CHK-101 | CHK-104 | Both ask for guest checkout without an account | CHK-101 states the reason (the forced sign-up step is the biggest funnel drop-off). Carry CHK-104's reporter over as a watcher |

## Proposed ordering (open, not yet started)

| # | Key | Title | Why here | Ready to pull? |
|---|---|---|---|---|
| 1 | CHK-107 | Update PCI self-assessment questionnaire | Fixed external date: submit before 31 Oct. Owned, 2 points, definition of done present | Yes |
| 2 | CHK-103 | Apple Pay button missing on Safari 17 | 14 customer reports; blocks a payment method. Owned, 3 points | Yes |
| 3 | CHK-119 | Accessibility fixes on the payment form | Screen readers skip a payment field, so some shoppers can't pay | No: needs an owner and acceptance criteria |
| 4 | CHK-112 | One-click reorder from order history | High, owned, acceptance criteria, 5 points | Yes |
| 5 | CHK-101 | Guest checkout without creating an account | Highest value claim in the backlog, but 13 points, unowned, 113 days without an update | No: split first, then assign |
| 6 | CHK-105 | Show delivery date estimate on the cart page | Medium, owned, acceptance criteria, 3 points | Yes |
| 7 | CHK-116 | Persist cart across devices for signed-in users | Medium, owned, acceptance criteria, 5 points | Yes |

In progress and left where they are: CHK-102, CHK-113.

## Priority conflicts to resolve

- **Five items at Highest (28% of the backlog).** CHK-120 (analytics dashboard) has had no update
  in 142 days and has no acceptance criteria. Proposal: drop it to Medium unless someone owns a
  date for it.
- **CHK-114 Klarna is Highest but unowned, 13 points, no acceptance criteria.** The description
  says a partner deal is signed. Ask whether the deal carries a launch date; if it does, it
  moves above CHK-112 once split.
- **CHK-106 payments rebuild is High, 21 points, 182 days untouched.** Keep High only if the
  team can name the first slice; otherwise move it to Low and revisit at planning.

## Change list

| Key | Change | Reason |
|---|---|---|
| CHK-117 | Close as duplicate of CHK-116 | Same request, see above |
| CHK-109 | Close as duplicate of CHK-103; copy customer note | Same symptom |
| CHK-104 | Close as duplicate of CHK-101 | Same story |
| CHK-101, CHK-114, CHK-106 | Split with `user-story-splitter` | Estimates 13, 13, 21 (limit 8) |
| CHK-101, CHK-114, CHK-119, CHK-111 | Assign an owner | Open with no assignee |
| CHK-119, CHK-110, CHK-120 | Add acceptance criteria | None found |
| CHK-120 | Priority Highest → Medium (proposal) | Stale 142 days, no criteria |
| CHK-115 | Close: won't do | Lowest, no owner, 324 days without an update |
| CHK-111, CHK-110 | Ask the reporter to confirm or close | 228 and 154 days without an update |

## Questions before applying

1. Who owns CHK-101, CHK-114 and CHK-119?
2. Does the Klarna deal (CHK-114) have a launch date?
3. Is CHK-120 still Highest, and for whom?

What the triage did not do: it didn't change any priority on its own, and it didn't invent an owner or a date. Each of those is a question in the last section.

How it's checked

Measured against plain Claude Code (2026-10-23): it met 75% of its checks, against 59% without it, and a blind judge preferred it. How skills are checked

26 assertions across 4 acceptance evals, and 8 trigger queries. Every skill's checks

What the output must do

Runs the triage script (or states the counts from the export) rather than estimating counts
Each duplicate names both keys, why they are the same work, and which one survives
Rejects or questions candidates that are similar in title but different work, rather than closing every flagged pair
Each item in the proposed ordering has a reason and says whether it is ready to pull
Oversized items are handed to user-story-splitter rather than scheduled as is
Priority changes are presented as proposals with the facts behind them, not applied silently
Ends with a change list of key, change and reason
Does not invent owners, dates, estimates or customer counts not in the export
Notes that the estimate and priority columns are missing and that the oversized and priority checks are skipped or mapped from labels
Uses --map or label-based priority rather than guessing priorities
Reports staleness from updatedAt with exact day counts
Confirms duplicates by reading descriptions, not title similarity alone
Does not invent a priority or estimate for items that have none
Identifies PARK-101 and PARK-117 as the same work, and does not merge the north and south gate benches (PARK-104/105) or the pothole patch and the full resurface (PARK-118/115)
The proposed two-week pull leaves out PARK-110 and PARK-112, which need the cherry picker, and adds up to no more than about 20 crew-days of estimated work
Flags that PARK-108 (exposed wiring) is at Low priority despite being a safety hazard, and proposes raising it rather than silently changing it
Does not give a total for the whole backlog as if it were known; reports the 37 crew-days across the 14 estimated items and says the rest have no estimate
Corrects the claim that nothing is older than six months, naming PARK-113 and PARK-115 (and PARK-114) as older
Does not invent owners, estimates or dates not in the export
Fits on about one page
Gives exactly five bullets
Treats the three panto costume entries as one item, not several
Puts the overdue fire extinguisher check in the top five, and ranks the dated grant (20 Nov) and raffle licence (1 Dec) above undated nice-to-haves like repainting the foyer or the website
Gives each item a short reason
Does not invent dates, owners or a panto date not in the list, and treats Ray as uncertain
Uses plain language with no tracker jargon

When it should and shouldn't activate

Should activate:

Triage our Jira backlog export and find the duplicates
We have 200 tickets in Linear, half of them stale. Which should we close and what should we pull next?
Groom this backlog CSV before sprint planning: missing owners, missing acceptance criteria, things too big to schedule
Find near-duplicate issues in our GitHub project and propose a priority order

Should leave to another skill:

A customer says checkout is broken on mobile sometimes. Turn this into a proper bug ticket
Break this guest-checkout epic into user stories with acceptance criteria
Write the weekly status report for the checkout team from this Jira export
Draft a PRD for guest checkout from this brief

Hands off to

user-story-splitter — when an item is over the size limit or too vague to estimate. Split it into independently shippable stories with acceptance criteria before it is scheduled.

Source

SKILL.md · the whole folder on GitHub.

backlog-triage