reflect

name	reflect
description	Daily and weekly review of the second brain. Use for /reflect, 'daily review', 'what's in my inbox', 'weekly review', 'what should I focus on'. Surfaces unprocessed captures, overdue tasks, stream health, detected connections, and MEMORY.md update suggestions.

Reflect Skill

Daily and weekly review. Surface what needs attention, find hidden connections, keep the vault clean.

Read the yaml config block in CLAUDE.md for the streams list and thresholds before starting.

When to Run

• Morning: "what's in my inbox" / daily review
• Evening: "log today's decisions"
• Weekly (suggested: Sunday): full weekly flow

---

Daily Reflect Flow

1. Inbox Scan

Check vault/inbox/ for unprocessed files.

• Count them and list titles
• For each: suggest classification and routing (a mini /capture proposal)
• Ask: "Process these now or later?"

2. Stream Health Check

Read all files in vault/streams/. For each active stream:

1. Parse frontmatter for nudge_after_days and priority
2. Find the last Activity Log entry; also check the index for recent items with matching stream_id
3. Calculate days since last activity
4. Assign health: Active (within threshold), Cooling (>50% of threshold elapsed), Starving (past threshold)

Stream Health:
STARVING  content-creation — 12d silent (threshold: 7). Goal: 2 posts/week → draft one from a recent insight?
COOLING   side-project — 5d (threshold: 7)
ACTIVE    day-job — today
ACTIVE    learning — 2d

For each starving stream: pull Active Goals from its stream file and suggest ONE concrete next action. One — not a list.

3. Tasks Status

Read vault/tasks/ files with status: To Do or In Progress.

• Flag anything with due_date today or past
• Flag p0_urgent / p1_important not yet started

Open tasks (3):
- [P0] Renew domain — due YYYY-MM-DD (OVERDUE)
- [P1] Review onboarding flow — no due date
- [P2] Read the caching article — no due date

4. Unsynced Captures (OPTIONAL — only if a cloud mirror or external channels are enabled)

If integrations.cloud_mirror is enabled: check the mirror for items captured via external channels (mobile/email/chat) not yet in the vault. List them and offer: "Sync these to the vault now?" If no mirror: skip.

5. Connection Detection

For items captured in the last 7 days, look for hidden connections:

• Notes sharing entities or stream context
• Ideas that support or contradict each other
• New items related to an existing hub page

Possible connections:
- "Referral program idea" ↔ "Growth experiments review" (both: organic growth)
  → add a wikilink between them?

6. MEMORY.md Update Proposals

Based on recent sessions, propose additions to vault/MEMORY.md — new preferences observed, patterns detected, decisions worth a durable record. Show proposals for confirmation before writing:

Proposed MEMORY.md additions:
- [YYYY-MM-DD] Prefers voice capture while commuting; these are usually ideas, not tasks.
  → Add? [yes/no]

---

Weekly Reflect Flow

Run the full daily flow PLUS:

7. Weekly Summary

• Total captures this week by type
• Most active streams
• Ideas that moved from New to In Progress

8. Pattern Detection

Across the last 7 days of captures:

• Topics mentioned 3+ times → candidate concept pages (hand to /lint)
• Streams with concentrated activity
• Entities appearing across multiple items

9. Stream Week-in-Review

| Stream | This week | Trend | Top goal progress |
|--------|-----------|-------|-------------------|
| day-job | 8 items | up | workshop prep done |
| side-project | 3 items | flat | launch checklist 60% |
| content-creation | 0 items | down | no posts this week |

Also list cross-stream opportunities (an insight in one stream that could feed another — e.g., "side-project growth data → content material").

10. Stale Items Check

• Ideas at status: New > 14 days → decide: develop, archive, or delete
• Tasks at To Do > 30 days → reschedule or close
• Active drafts > 48h → expire?
• MEMORY.md facts [verified:] > 90 days ago → re-verify?

11. Classification Accuracy

Read vault/notes/accuracy-log.md for the last 7 days; count total classifications from vault/log.md.

Classification accuracy:
- 23 items classified this week, 1 correction (idea → task) → 95.7%

If accuracy ≥ 90% and volume ≥ 20 over a rolling 14 days, the capture skill is eligible for Partner autonomy (criteria in vault/AGENTS.md):

• Propose: "Capture skill has met Partner criteria. Promote? [yes/no]"
• On yes: update the vault/AGENTS.md Skill Registry table and log the promotion in vault/MEMORY.md under Autonomy Promotions.

If accuracy < 90%: note which types are confused most often and propose a classifier-rule fix (via the /correct recurrence pattern).

12. Improvement Candidates

Propose at most 3 concrete system improvements observed this week (awkward flows, missing links, rule gaps). File accepted ones to vault/notes/ as retro notes — with the full compile step.

---

Wrap-up

Every reflect run ends with:

1. A vault/log.md line: ## [YYYY-MM-DD HH:MM] reflect | daily (or weekly) — N inbox, N nudges, N proposals
2. The question: "What would you like to do first?"

Keep output tight — compact lists, not paragraphs. The owner wants decisions, not summaries.