Next Priority Tasks

Aggregate your tracked-work sources into a single ranked list of the top 10 actionable items, each with a proposed action and reasoning. The author's own setup has three sources — system logs, UX assignments, and filed proposals, each backed by a private admin API — but the method generalizes to whatever you actually have: even a single source (say, just /how-to-submit-and-track-proposals's local proposal folder, or a plain TODO list) still benefits from being scored and ranked the same way.

Is the private admin API set up?

Each data source below has two paths: API (if you've set up the person's private admin backend — see /alignment-harness:harness-setup) and local fallback (works immediately, no setup). Check once at the start of a session and use whichever applies; don't half-query one and half the other for the same source.

Atomic Skills This Composes (optional — use if you have them)

Skill	Role in Workflow
system-log-triage (if you have it)	Defines severity scale, terminal statuses, decision tree for error logs
ux-assignment-generator (if you have it)	Defines UX assignment statuses, item model, scoring
how-to-submit-and-track-proposals	Defines proposal statuses, priority scale, and both the API and local-folder shape

If you don't have the first two, treat "system logs" and "UX assignments" below as illustrative categories you can skip, or replace with whatever equivalent tracked-work sources your own project has (a bug tracker, a QA checklist, a monitoring dashboard's alert feed — anything with a status and a severity/priority).

Data Sources

1. System / error logs (severity-sorted errors, last 7 days)

API (if set up): query your admin backend's error-log endpoint for non-terminal errors, severity 5+, from the last 7 days, sorted by severity descending, with an API key read from your own env config — never hardcode the key or the host in a skill file.

Local fallback: check whatever your project already uses to surface errors — a Sentry/error-tracker export, an errors.log file, or alignment-harness records system-logs if you've been logging them there yourself. If none exists, skip this source and say so.

Terminal statuses to EXCLUDE (already resolved):

• dismissed (closed, will reopen if error recurs)
• silenced (permanently hidden)
• solution implemented AND reviewed (fix confirmed)

Non-terminal statuses to INCLUDE (needs action):

• No solution proposed (needs investigation)
• Solution proposed but not accepted (needs human review)
• Solution accepted but not implemented (needs agent work)
• Solution implemented but not reviewed (needs human verification)

Priority mapping: severity 8-10 = CRITICAL, 5-7 = HIGH

2. UX / QA assignments (non-terminal, with admin engagement)

API (if set up): query your admin backend for assignments where the admin commented and/or explicitly requested agent help, plus items still pending review.

Local fallback: check alignment-harness records ux-assignments if you keep these locally, or your project's own QA/issue tracker for open items tagged for agent attention.

Terminal statuses to EXCLUDE: verified, failed, skipped, stale

Non-terminal statuses to INCLUDE: pending, proposed, deferred, agent-input, on-hold

Priority boost: Items with status agent-input get +30 priority (the admin is blocked waiting for the agent)

3. Agent proposals (high-priority, non-terminal, or help-requested)

API (if set up): query your admin backend for proposals needing an agent response (the admin asked a question), high-priority approved proposals ready for implementation, and pending proposals above your own priority threshold.

Local fallback: run alignment-harness records proposals to get the local folder (see /how-to-submit-and-track-proposals) and read the proposal files there for status and priority.

Terminal statuses to EXCLUDE: implemented, rejected

Non-terminal statuses to INCLUDE: pending-review, approved, in-progress, deferred, agent-input

Priority boost: agent-input status gets +30 (the admin is waiting)

4. Your task tracker (always available, no setup)

If you're using Claude Code's own task tracking (TaskList/TaskGet) or any other lightweight tracker, include its open items as a fourth source — this one never needs an API and always works, so it's worth including even when nothing else is set up.

Unified Priority Score (0-100)

Normalize all items to a single 0-100 scale:

Source	Score Formula
System log (sev 8-10)	90 + (severity - 8) * 3
System log (sev 5-7)	60 + (severity - 5) * 10
UX assignment (agent-input)	80
UX assignment (pending, high importance)	importanceScore * impactScore / 10 (capped at 70)
Proposal (agent-input)	85
Proposal (approved)	proposal.priority
Proposal (pending, priority >= 80)	proposal.priority - 10
Task tracker item (no other score available)	use its stated priority, or default to 50

Tie-breaker: More recent items win (sort by last-seen or last-updated timestamp; fall back to creation date)

Presentation Format

For each of the top 10 items, present:

## #{rank} [{source}] {title} — Score: {score}

**Source**: System Log / UX Assignment / Proposal / Task
**Status**: {current status} | **Since**: {date}
**Why it's priority #{rank}**: {1 sentence — what makes this urgent and who is blocked}

**Proposed Action**: {specific next step the agent or human should take}
**Effort**: {quick estimate: trivial / small / medium / large}
**Where to see it**: {a clickable admin URL if you have one set up, otherwise the local file path or task ID}

Where to see it, by source

Source	If you have an admin UI	Local fallback
System Log	link into your own admin UI's log view	the log file/entry path
UX Assignment	link into your own admin UI's assignments view	the local record's file path
Proposal	link into your own admin UI's proposals view	the proposal's file path under alignment-harness records proposals
Task	(task trackers don't usually have a URL)	the task ID

Workflow Steps

1. Fetch — Query every source you actually have (API and/or local), in parallel where possible
2. Filter — Remove terminal items, remove system logs below severity 5
3. Score — Apply the unified priority score formula
4. Dedupe — If a system log has a linked proposal, show only the higher-scored one
5. Sort — By score desc, then recency
6. Present — Top 10 in the format above
7. Ask — "Want me to start on #1, or do you want to pick a different one?"

Anti-Patterns

• Do NOT show terminal items (already done)
• Do NOT show system logs below severity 5 when severity 5+ exist
• Do NOT show proposals below priority 60 unless they have agent-input status
• Do NOT propose actions that require human approval without saying so
• Do NOT skip the "proposed action" — every item needs a concrete next step
• Do NOT silently skip a source because its API isn't configured — say plainly which sources you checked and which you couldn't (and why), so the person knows the list isn't exhaustive