SKILL.md
Backlog triage skill
An evidence-first workflow to deduplicate and prioritize work using impact, evidence, dependency, and risk.
- Revision
- 2
- Verified
- 2026-10-20
Compatibility and paths
Codex
skills/backlog-triage/SKILL.mdClaude Code
.claude/skills/backlog-triage/SKILL.mdVS Code
.github/skills/backlog-triage/SKILL.mdTrust and provenance
Curated record reviewed 2026-10-20. Results still depend on the supplied context and target environment.
Generated assetReady to copy or download
---
name: backlog-triage
description: Produces a dependency-aware next-action queue while preserving uncertain duplicates and completion evidence. Use when tracker status, repeated titles or blocked work make ordering unclear; propose changes without silently merging or closing issues.
license: CC-BY-4.0
compatibility: Complete fictional nine-issue ledger uses Node.js 22.13+ only if run. No tracker, account, credentials or network required. Real issue mutations/communications need the task owner's authority.
metadata:
author: oneliners
version: "2.0.0"
---
# Backlog triage skill
## Read the objective, not just the largest count
Identify current goal, owner ordering, affected user/task, evidence freshness and delivery/acceptance constraints. Separate incidents, blockers, implementation, discovery and later ideas. Preserve explicit owner holds/cancellations. An old done status or a popular title alone does not establish either a delivered fix or a duplicate.
## Complete original issue-ledger case
The nine synthetic issues below contain every status, owner-supplied rank and dependency needed by this exercise. verifiedDone is this exercise's explicit acceptance-evidence flag, not a required property of any real tracker. Matching strings produce candidates only.
Save the full next block as backlog-fixture.mjs and run node backlog-fixture.mjs with Node.js 22.13+. It only analyzes embedded JSON in memory; it does not connect to a tracker or change a task.
```js
import assert from "node:assert/strict";
const triageIssues = [
{
"id": "DEMO-1",
"title": "Export fails",
"component": "export-api",
"version": "1.8",
"environment": "staging",
"symptom": "401 after session refresh",
"state": "todo",
"verifiedDone": false,
"ownerRank": 2,
"requires": [
"DEMO-4"
],
"evidence": "owned request trace e1"
},
{
"id": "DEMO-2",
"title": "Download does not start",
"component": "export-api",
"version": "1.8",
"environment": "staging",
"symptom": "401 after session refresh",
"state": "todo",
"verifiedDone": false,
"ownerRank": 3,
"requires": [],
"evidence": "second owned trace e2; root cause not yet compared"
},
{
"id": "DEMO-3",
"title": "Export fails",
"component": "mobile-download-ui",
"version": "1.7",
"environment": "staging",
"symptom": "save dialog cancelled",
"state": "todo",
"verifiedDone": false,
"ownerRank": 5,
"requires": [],
"evidence": "owned mobile case e3"
},
{
"id": "DEMO-4",
"title": "Reproduce refresh handoff",
"component": "session-boundary",
"version": "1.8",
"environment": "staging",
"symptom": "need exact credential handoff",
"state": "todo",
"verifiedDone": false,
"ownerRank": 1,
"requires": [],
"evidence": "bounded trace task e4; blocks DEMO-1"
},
{
"id": "DEMO-5",
"title": "Explain cancelled downloads",
"component": "docs",
"version": "1.7",
"environment": "staging",
"symptom": "missing user guidance",
"state": "todo",
"verifiedDone": false,
"ownerRank": 6,
"requires": [],
"evidence": "doc gap e5"
},
{
"id": "DEMO-6",
"title": "Compatibility gate",
"component": "release",
"version": "1.8",
"environment": "staging",
"symptom": "done label has no acceptance evidence",
"state": "done",
"verifiedDone": false,
"ownerRank": 4,
"requires": [],
"evidence": "status only; proof not supplied"
},
{
"id": "DEMO-7",
"title": "Enable reviewed integration",
"component": "release",
"version": "1.8",
"environment": "staging",
"symptom": "await compatibility evidence",
"state": "todo",
"verifiedDone": false,
"ownerRank": 7,
"requires": [
"DEMO-6"
],
"evidence": "acceptance depends on verified gate"
},
{
"id": "DEMO-8",
"title": "Agree ownership",
"component": "process",
"version": "draft",
"environment": "planning",
"symptom": "circular prerequisite",
"state": "todo",
"verifiedDone": false,
"ownerRank": 8,
"requires": [
"DEMO-9"
],
"evidence": "dependency ledger"
},
{
"id": "DEMO-9",
"title": "Approve ownership map",
"component": "process",
"version": "draft",
"environment": "planning",
"symptom": "circular prerequisite",
"state": "todo",
"verifiedDone": false,
"ownerRank": 9,
"requires": [
"DEMO-8"
],
"evidence": "dependency ledger"
}
];
function reviewBacklog(issues) {
const byId = new Map();
for (const issue of issues) {
assert.ok(issue.id && !byId.has(issue.id), "unique issue ID required");
assert.ok(["todo", "active", "done", "cancelled"].includes(issue.state) && Array.isArray(issue.requires) && Number.isInteger(issue.ownerRank), "explicit status/dependencies/owner rank required");
byId.set(issue.id, issue);
}
const missingDependencies = issues.flatMap((issue) => issue.requires.filter((id) => !byId.has(id)).map((id) => ({ issue: issue.id, missing: id })));
const cycles = [], visited = new Set(), stack = [];
function visit(id) {
const position = stack.indexOf(id);
if (position >= 0) { cycles.push([...stack.slice(position), id]); return; }
if (visited.has(id) || !byId.has(id)) return;
stack.push(id); for (const next of byId.get(id).requires) visit(next); stack.pop(); visited.add(id);
}
for (const id of byId.keys()) visit(id);
const completed = (id) => byId.get(id)?.state === "done" && byId.get(id)?.verifiedDone === true;
const cycleMembers = new Set(cycles.flat());
const workable = issues.filter((issue) => issue.state === "todo" && !cycleMembers.has(issue.id) && issue.requires.every(completed)).sort((a, b) => a.ownerRank - b.ownerRank).map((issue) => issue.id);
const blocked = issues.filter((issue) => issue.state === "todo" && (cycleMembers.has(issue.id) || !issue.requires.every(completed))).map((issue) => ({ id: issue.id, requires: issue.requires.filter((id) => !completed(id)), cycle: cycleMembers.has(issue.id) }));
const candidateDuplicates = [];
for (let left = 0; left < issues.length; left++) for (let right = left + 1; right < issues.length; right++) {
const a = issues[left], b = issues[right];
if (["component", "version", "environment", "symptom"].every((key) => typeof a[key] === "string" && a[key].length > 0 && a[key] === b[key])) candidateDuplicates.push([a.id, b.id]);
}
return { workable, blocked, missingDependencies, cycles, candidateDuplicates, completionEvidenceReview: issues.filter((issue) => issue.state === "done" && issue.verifiedDone !== true).map((issue) => issue.id), scope: "Local ledger consistency and owner-supplied ordering, not automatic priority, merge, completion or tracker changes." };
}
const result = reviewBacklog(triageIssues);
assert.deepEqual(result.workable, ["DEMO-4", "DEMO-2", "DEMO-3", "DEMO-5"]);
assert.deepEqual(result.candidateDuplicates, [["DEMO-1", "DEMO-2"], ["DEMO-8", "DEMO-9"]]);
assert.deepEqual(result.completionEvidenceReview, ["DEMO-6"]);
assert.deepEqual(result.cycles, [["DEMO-8", "DEMO-9", "DEMO-8"]]);
console.log(JSON.stringify(result, null, 2));
```
## Worked triage and decisions that remain human
Next bounded action is DEMO-4, according to the supplied owner rank: gather the refresh-boundary evidence that blocks DEMO-1. Workable is DEMO-4, DEMO-2, DEMO-3, DEMO-5; this list is a dependency/status screen, not permission to implement every item at once. DEMO-2 should first be compared with DEMO-1, not generate a second uncoordinated fix.
DEMO-1/2 have matching component/version/environment/symptom but only candidate duplicate status; compare traces, reproduction and acceptance before proposing a canonical issue/related link. The identical title of DEMO-1/3 is insufficient: mobile save cancellation on 1.7 differs from an API 401 after refresh on 1.8. DEMO-8/9 also look similar to a string matcher, but are distinct mutually dependent decisions: the cycle DEMO-8->DEMO-9->DEMO-8 must be resolved with the owner, not automatically merged.
DEMO-6 says done but lacks verified acceptance in this supplied ledger. DEMO-7 stays blocked pending the missing evidence. Ask for the actual commit/release/test/acceptance proof; do not automatically reopen real completed work merely because an optional field is absent in a different tracker. Cancelled and done are not interchangeable completion semantics, and deleting a dependency to make a card workable does not deliver it.
Keep priorities tied to user impact/current objective and evidence confidence. The ranks here are explicitly provided, not computed from guesses about users, severity or effort. Return a small next-action queue, blocker graph/cycle, completion-evidence questions and candidate duplicate links with rationale. Do not close/merge/delete/assign issues or message stakeholders on the basis of this local report. No GitHub/Lific operation occurred.
## Verify the relationships before a mutation
[GitHub dependency documentation](https://docs.github.com/en/issues/tracking-your-work-with-issues/using-issues/creating-issue-dependencies) illustrates blocker direction; apply the actual semantics of the team's chosen tracker rather than assuming all status labels match. Confirm a duplicate's same problem/version/context and preserve history/links if the owner approves consolidation. Resolve cycles or missing references visibly; do not call them completed. Record proposed priority, reason, blocker/decision owner, smallest next step and acceptance evidence. Missing impact/authority or conflicting owner instructions stays an explicit question, not a fabricated numerical score.
Real example
Input
Review the complete nine-issue DEMO ledger in SKILL.md. Respect owner ranks/dependencies, distinguish title similarity from duplicate evidence, identify the DEMO-8/9 cycle and missing acceptance proof for a done gate. Produce proposals only; do not connect to or mutate a tracker.
Expected result
Next bounded action is DEMO-4, according to the supplied owner rank: gather the refresh-boundary evidence that blocks DEMO-1. Workable is DEMO-4, DEMO-2, DEMO-3, DEMO-5; this list is a dependency/status screen, not permission to implement every item at once. DEMO-2 should first be compared with DEMO-1, not generate a second uncoordinated fix. DEMO-1/2 have matching component/version/environment/symptom but only candidate duplicate status; compare traces, reproduction and acceptance before proposing a canonical issue/related link. The identical title of DEMO-1/3 is insufficient: mobile save cancellation on 1.7 differs from an API 401 after refresh on 1.8. DEMO-8/9 also look similar to a string matcher, but are distinct mutually dependent decisions: the cycle DEMO-8->DEMO-9->DEMO-8 must be resolved with the owner, not automatically merged. DEMO-6 says done but lacks verified acceptance in this supplied ledger. DEMO-7 stays blocked pending the missing evidence. Ask for the actual commit/release/test/acceptance proof; do not automatically reopen real completed work merely because an optional field is absent in a different tracker. Cancelled and done are not interchangeable completion semantics, and deleting a dependency to make a card workable does not deliver it. Keep priorities tied to user impact/current objective and evidence confidence. The ranks here are explicitly provided, not computed from guesses about users, severity or effort. Return a small next-action queue, blocker graph/cycle, completion-evidence questions and candidate duplicate links with rationale. Do not close/merge/delete/assign issues or message stakeholders on the basis of this local report. No GitHub/Lific operation occurred.