follow-up-patrol

${var} — Optional filter (e.g. "security", "release", "disclosure"). If empty, audits everything.

Today is ${today}. This skill watches the follow-up backlog so nothing falls through the cracks between self-reviews.

Voice

If soul/SOUL.md and soul/STYLE.md exist and are populated, read them and match the operator's voice in the notification. Otherwise use a clear, direct, neutral tone — state what needs to happen, no hedging.

Why this skill exists

Open loops accumulate in MEMORY.md between reviews. Items requiring a human action can sit for weeks without anything escalating them by age. Other skills cover specific lanes (e.g. pr-tracker for bot PRs, disclosure-tracker for pending disclosures, pvr-watchlist for enablement probes). This skill is the catch-all — the general follow-up backlog plus the issue tracker, escalated by age.

Steps

1. Read the follow-up section

Read memory/MEMORY.md. Find the follow-up section using this resolution order:

1. A ## Known Follow-ups section, if present (the canonical home).
2. Otherwise ## Next Priorities (the default section a fresh Aeon ships with).
3. Otherwise ## Open Loops or any heading containing "follow" / "todo" / "pending".

Extract every bullet point under the matched section. If none of these sections exist, skip follow-up parsing and proceed with the issue tracker only (step 3).

For each bullet, parse:

• Label — the name/identifier before the em dash or first —
• Date mentioned — look for dates like 5/12, 5/25, since 4/30, 24d+, 12d, an ISO date, etc.
• Status type — classify as one of:
  • operator — requires manual human action (e.g. "operator: open PR", "rotate key X")
  • auto — being handled by another skill on a schedule (e.g. "pvr-watchlist probing 6/1")
  • deferred — explicitly deferred pending a decision
  • deadline — has a hard date (e.g. "2026-06-12")
• Description — the full bullet text, one line

2. Compute age for each item

Today = ${today}. For each item:

• If it contains a date like 5/12 or since 5/12 → age = days since that date
• If it says 24d+ → age = 24 (or more)
• If it says 12d → age = 12
• If no date is detectable → age = unknown (treat as WATCH)

Apply urgency tiers:

Tier	Condition	Action
CRITICAL	age > 21d AND type = operator	Must escalate immediately
HIGH	age > 14d AND type = operator	Escalate this cycle
MEDIUM	age > 7d (any type)	Note and monitor
WATCH	age ≤ 7d OR type = auto/deferred/deadline	Informational only

If ${var} is set, only process items whose text contains ${var} (case-insensitive).

3. Read open issues

Read memory/issues/INDEX.md if it exists. Find the ## Open table. Extract each open issue row:

• ID, title, severity, detected date
• Compute age from detected date

Add them to the escalation picture (separate section in output). If memory/issues/INDEX.md doesn't exist, skip this section.

4. Write memory/topics/follow-up-status.md

Overwrite the file:

# Follow-up Status — ${today}

*Generated by follow-up-patrol. Next run: next Tuesday 11:00 UTC.*

## CRITICAL (operator action required, >21d)

| Item | Age | Type | Action Needed |
|------|-----|------|---------------|

## HIGH (operator action required, >14d)

| Item | Age | Type | Action Needed |
|------|-----|------|---------------|

## MEDIUM (>7d)

| Item | Age | Type | Notes |
|------|-----|------|-------|

## WATCH (<7d or auto-tracked)

| Item | Age | Type | Notes |
|------|-----|------|-------|

## Open Issues

| ID | Title | Severity | Age |
|----|-------|----------|-----|

---
*Items in CRITICAL/HIGH tiers require operator attention. AUTO items are handled by scheduled skills.*

5. Decide whether to notify

• If zero CRITICAL and zero HIGH items: skip notification. Log FOLLOW_UP_PATROL_SKIP: no escalation-tier items found.
• Otherwise: always notify.

6. Send notification

Write to .pending-notify-temp/follow-up-patrol-${today}.md (create the dir if needed):

follow-up patrol — ${today}

open issues: N (severities)

CRITICAL (>21d, operator): N items
- [label] — Nd — [one-line action]

HIGH (>14d, operator): N items
- [label] — Nd — [one-line action]

MEDIUM (>7d): N items
[brief list]

WATCH: N items (auto-tracked)

full breakdown: memory/topics/follow-up-status.md

Keep under 800 chars. State what needs to happen.

Then run:

./notify -f .pending-notify-temp/follow-up-patrol-${today}.md

7. Log to memory/logs/${today}.md

Append:

## Follow-up Patrol
- **Items parsed:** N
- **CRITICAL:** N
- **HIGH:** N
- **MEDIUM:** N
- **WATCH:** N
- **Open issues:** N
- **Notification:** sent / skipped
- FOLLOW_UP_PATROL_OK

If skipped: FOLLOW_UP_PATROL_SKIP: <reason>.

Required Env Vars

None. All reads from local memory/ files.

Sandbox Note

No network calls required. All data comes from local memory/MEMORY.md and memory/issues/INDEX.md — both committed to the repo and available in every run after checkout. The only outbound call is ./notify, which is already sandbox-safe.