quick-wins - SKILL.md Agent Skill

name: quick-wins description: Surface and triage local issues that are safe and quick for an AI to implement and easy for a human to verify. Use when the user wants to "knock out issues", asks for "quick wins", "low-hanging fruit", "what's easy to do right now", or wants a batch of small tasks to work through in one session. allowed-tools: Bash, Read

Quick Wins — find AI-friendly issues to knock out

Use this skill when the user wants to power through a batch of small issues. The goal is to identify issues where:

The spec is already done — root cause, file paths, or implementation hints are in the issue body
The blast radius is small — touches one logger / one command / one screen, not architecture
A human can verify in <2 minutes — clear "do X, see Y" check
Failure mode is obvious — if it breaks, you'll see it immediately (not a slow data corruption)

Workflow

Step 1 — list candidate issues

Pull all 2-todo issues with their size/complexity:

obsidian base:query vault=nvALT path="Simple History issues.base" format=json \
  | jq '[.[] | select(.status == "2-todo") | {name, prio, type, size, complexity}]'

Filenames also act as a length proxy (longer file = more spec written down):

ls -S "/Users/bonny/Documents/nvALT/Simple History/issues/" | head -30
# Or sort by line count:
wc -l "/Users/bonny/Documents/nvALT/Simple History/issues/"*.md | sort -n

Very short issues (<25 lines) are usually stubs without enough context — skip them unless the title alone is enough. Very long issues are often well-spec'd but may have grown in scope — read carefully.

Step 2 — score each candidate

Read the body and rate against these signals. Don't invent scores; just check off what's true.

Green flags (more = better)

🟢 Issue body names a specific file + line for the fix (e.g. "class-post-logger.php:654, add ...")
🟢 Has a > [!agent] decision block — means the design has already been debated
🟢 Marked size: 1-small or complexity: patch
🟢 Type is bug with reproducible steps
🟢 Verification is "run X, see Y" — no test data, no special account, no third-party service
🟢 Single-file or single-method change

Red flags (any one is usually disqualifying)

🔴 Touches the database schema or runs ALTER TABLE on existing data
🔴 Modifies billing, license validation, or update server logic
🔴 Crosses core ↔ premium boundary (without clear interface)
🔴 Type is feature and the body is a stub (<25 lines, no implementation hints)
🔴 Verification needs WooCommerce + a real order + a real product (unless WC is set up locally)
🔴 Status is 4-needs-decision or 5-needs-investigation — by definition not ready
🔴 Networks/multisite changes without a network test environment
🔴 UI redesigns (subjective taste calls — save for human-led work)

Step 3 — present a tiered list

Group findings into three tiers and present to the user. Don't pick — let the user choose what to work on.

🟢 Tier 1 — Spec is done, just implement
- <Issue name> — one-line summary of fix + verification step
...

🟡 Tier 2 — Clear scope, slightly larger
- <Issue name> — ...

🔴 Skip for now (with reason)
- <Issue name> — too risky / underspecified / needs decision

Step 4 — once user picks one

Follow the normal local-issues workflow:

obsidian property:set ... status=1-in-progress
Implement the change
Verify by running the steps in the issue
Run linters and relevant tests
Commit (per git-commits skill)
obsidian property:set ... status=8-done and review=pending
Append a > [!agent] block summarising what changed and how the user can verify

If the user says "do them all" or "knock them out", work through one at a time, marking each done before starting the next. Do not batch into a single commit — one issue, one commit, so review can happen per-issue.

Anti-patterns — DON'T

❌ Don't promote an issue to Tier 1 just because it's small. "Small + underspecified" still means hidden decisions.
❌ Don't treat size: 1-small as the only signal — read the body.
❌ Don't pick issues with status: 4-needs-decision or 5-needs-investigation — they're parked for a reason.
❌ Don't pick UX/redesign work without an explicit > [!agent] decision in the body.
❌ Don't assume an issue is fresh — check if there's already an agent note saying it's partly done.

Notes on common issue types in this repo

WP-CLI bugs (e.g. things-not-logged-from-CLI) → usually a missing WP_CLI exception alongside the REST_REQUEST/XMLRPC_REQUEST checks. Easy to verify with wp commands.
Logger improvements (extra context fields like SKU/ID) → see the create-logger and logger-messages skills. Verify via the populate-log skill or by doing the action and reading the event details.
License/constant features (premium) → small, isolated changes in the licensing code. Verify by define()-ing the constant and reloading the settings screen.
WooCommerce features → only safe if WC is installed locally and you can hit the action. If not, mark blocked and move on.