Triage Task

Triage Task

An advisory rubric for deciding what's worth an agent's time. This skill never mutates repo state — no labels, no assignees, no comments. It only produces a recommendation for a human or another agent to act on.

Inputs

gh issue list --label "status: available" --state open --json number,title,labels,body
gh issue list --label "status: changes-requested" --assignee "@me" --state open --json number,title,labels

Scoring

For each issue, score three axes:

Priority (1-5). Read from the priority: label if present, otherwise estimate from urgency described in the issue body. priority: high should generally dominate the final verdict regardless of the other two axes.

Value (composite). Consider:

• Impact if this ships (how many users/workflows does it affect?)
• Leverage (does it unblock other queued work, or is it a dead end?)
• Novelty (is this duplicate/overlapping with another open issue or a

recently merged PR? Check before scoring — duplicated effort is zero value regardless of how good the individual PR is.)

• Tractability (is the ask well-specified enough to execute without a lot

of back-and-forth, or is it actually a research/discovery task in disguise?)

Token-cost band (S/M/L/XL). Rough estimate of how much agent context and how many turns this will take: S = single-file, well-specified fix. M = a few files, some design judgment. L = cross-cutting change or non-trivial investigation. XL = probably needs to be broken into smaller issues before anyone should pick it up — flag this in your output rather than triaging it as-is.

Verdict

Combine into one of:

• 🟢 Do now — high priority or high value, tractable, reasonable cost.
• 🔵 Good ROI — solid value relative to cost, not urgent.
• 🟡 Defer — low urgency, or value is unclear until something else

lands first.

• 🔴 Skip — duplicate, out of scope, or cost (XL, ill-specified) far

exceeds likely value as currently written — recommend the issue be split or clarified instead of worked on directly.

Output format

A short table: issue number, one-line title, priority, value note, cost band, verdict. End with your top 1-3 recommendations for what to actually work on next, and a one-line reason for each.

What this skill must never do

• Never call gh issue edit, add/remove labels, change assignees, or

comment on issues as a side effect of triage.

• Never claim work on the triager's behalf — that's start_work.sh's job,

driven by the queue order in AGENT_CONTRACT.md, not by this skill's opinion.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: vanterx
• Source: vanterx/mssql-performance-skills
• License: MIT

Install and usage instructions live in the source repository linked above.