review

Read every context.md and reference.md in the repo. These are the review's data sources.

Discover, don't enumerate. Do not work from a hard-coded list — it goes stale as domains are added, and the review's whole job is catching staleness. Discover the files at runtime:

# Tier 1 — working memory (always read)
find . -name context.md -not -path './.git/*' | sort
# Tier 2 — reference memory (read for depth, skip with --quick)
find . -name reference.md -not -path './.git/*' | sort

Always read every context.md the first command returns, plus decisions.md (root). Read them in parallel — they're small. This guarantees no domain is ever missed.

Also read (Tier 2 — for depth, when not using --quick): every reference.md the second command returns.

If --domain flag is set, only report on that domain but still read root files (context.md, decisions.md) for cross-referencing.

2. Staleness Audit

For each domain, assess freshness:

Session History Check

• Parse the most recent session history entry date from each context.md
• Calculate days since last session activity
• Categorize:
  • Active: Session within last 14 days
  • Cooling: 15-30 days since last session
  • Stale: 31-90 days since last session
  • Dormant: 90+ days since last session

Content Freshness

• Check "Last updated" dates in context files
• Flag any section explicitly marked with a date that's >60 days old
• Check for items marked "TBD", "TODO", "Action needed", "Not started" — these are open loops

Context Size Check

• Count lines in each context.md
• Flag any >100 lines (may need archiving or tier split to reference.md)

Session History Cap (protocols/memory.md)

• Flag any context.md whose session history exceeds ~5 entries or ~1 month — older entries graduate to {domain}/archive/sessions-{year}.md during the interactive phase
• Flag option-comparison working docs (palettes, sourcing comparisons, shortlists) whose decision has been made — they graduate to {domain}/archive/ (Archive After Decision rule)

3. Open Loops Detection

Scan all context files for unresolved items:

Action Items

Look for patterns indicating incomplete work:

• "Not started" / "not yet" / "TBD" / "TODO"
• "Action needed" / "needs" / "research needed"
• Unchecked items in checklists (- [ ])
• "Waiting on" / "blocked by"
• Items with target dates that have passed

Stale Decisions

From decisions.md:

• Decisions with review dates that have passed
• Decisions with review dates within 30 days
• Decisions whose assumptions may have changed (cross-reference with context)

Orphaned Priorities

Items listed as "active priorities" in context files that haven't shown progress in session history.

4. Generate Reflection Questions

Based on what was found, generate 3-5 targeted reflection questions. These should be:

• Specific — based on actual context, not generic
• Forward-looking — "What's the next step on X?" not "How do you feel about X?"
• Cross-domain — spot connections the user might miss
• Actionable — should lead to a concrete update or decision

Question Templates (adapt to findings)

For stale domains:

"{Domain} hasn't been touched in {N} days. Has anything changed, or is it stable by design?"

For open loops:

"You flagged {item} as needed back in {date}. Still relevant, or can we close this?"

For approaching deadlines:

"{Deadline} is {N} days away. What's the status on {related item}?"

For cross-domain connections:

"The {domain1} plan depends on {assumption from domain2}. Has that assumption held up?"

For decisions due review:

"Decision {ID} was set for review {trigger}. The context now is {summary}. Worth revisiting?"

For life-stage transitions:

"With {life event} approaching, have you thought about how it affects {domain}?"

5. Domain-by-Domain Status

For each domain with activity (skip truly dormant domains unless they have open loops):

### {Domain Name}
**Last active**: {date} ({N days ago})
**Status**: {Active / Cooling / Stale / Dormant}

**What's current:**
- {Key current items from context}

**Open loops:**
- {Any unresolved items}

**Suggestion:** {One actionable suggestion if relevant}

Ordering

1. Domains with open loops or approaching deadlines first
2. Active domains next
3. Stale/dormant domains last (brief mention only)

6. Format the Review

Header

# Weekly Review: {Day of Week}, {Date}
**Leo**: {age in days/weeks/months} | **Paternity leave ends**: {N} days (Aug 1)

Section: Health Check

## System Health

| Domain | Last Active | Status | Open Loops |
|--------|-------------|--------|------------|
| Baby | Feb 5 | Active | 2 |
| Finances | Feb 3 | Active | 1 |
| Career | Feb 1 | Cooling | 3 |
| ... | ... | ... | ... |

**Context files**: {X} of {Y} under 100 lines ✓
**Decisions**: {N} active, {M} due for review

Section: Needs Attention

## Needs Attention

### Open Loops
- **{Domain}**: {Item} — {context}
- **{Domain}**: {Item} — {context}

### Decisions Due for Review
- **{ID}: {Title}** — {Review trigger}. {Brief context on why it matters now}

### Approaching Deadlines
- **{Date}** — {Item} ({domain})

Section: Reflection Questions

## Reflection Questions

1. {Question}
2. {Question}
3. {Question}

Present these as a conversation — pause after questions and wait for the user to respond. Don't rush past them.

Section: Domain Status (full review only, skip with --quick)

## Domain Status

{Domain-by-domain breakdown from step 5}

Footer

---
*Take your time with the questions. We can update context files with anything that's changed.*

7. Interactive Phase

After presenting the review, shift to interactive mode:

1. Wait for responses to reflection questions
2. Update context files based on answers
3. Close open loops — mark items as done, remove stale ones, update status
4. Log decisions if any are made during the review (offer to use decision journal)
5. Enforce the Session History Cap — graduate entries older than ~5 sessions/1 month to {domain}/archive/sessions-{year}.md, and decided option-docs to archive (see protocols/memory.md anti-bloat rules)

This is a conversation, not a report. The review is the prompt for a productive discussion.

8. Update Session History

After the review is complete, add an entry to projects/ai-assistant/context.md:

### YYYY-MM-DD
- Ran weekly review
- Domains reviewed: {count}
- Open loops found: {count}
- Items resolved: {count}
- Context files updated: {list}

Also update session history in any domain context files that were modified during the review conversation.