| name | weekly-wrap |
| description | Weekly review + compound learnings — review progress, surface blockers, distill insights into Memory |
Weekly Wrap
Run a structured weekly review, produce a shareable status update, then distill learnings into compounded knowledge.
Instructions
When the user invokes /weekly-wrap, run all steps in sequence without waiting for prompts between them. Produce the final shareable update, learnings, and next-week focus in a single pass. Save everything, then confirm.
Do NOT stop to ask for confirmation between steps. The user will correct anything after seeing the full output.
Step 1: Gather the Week's Raw Material
Pull everything from the target week before synthesising anything.
- Weekly scratchpad — read
Tasks/Week-YYYY-WNN.md (ISO week number, e.g. Tasks/Week-2026-W11.md)
- Count ticked
[x] vs unticked [ ] items across all daily sections (check ### Do Next and ### Follow-ups sections; for older weeks also check ### Planned, ### Deep Work, ### Quick Wins)
- Extract notes, decisions, and meeting outcomes from
### Log sections (for older weeks also check ### Decisions & context, ### Blockers, ### Notes)
- Note which days had entries
- Meeting notes — scan
Context/Meeting Notes/ for files dated this week
- Extract: attendee names, decisions made, action items with owners, blockers raised
- Task files — read
Tasks/ root for status: d (done this week) and status: b (blocked)
- Also scan
Tasks/Backlog/ to assess initiative-level progress beyond individual task completions — flag any track that has had no slice activity this week
- Also flag
status: s with no progress log entry in the last 7 days (stalled)
- Goals — read
GOALS.md for quarterly objectives and top 3 priorities
- Prior wrap — read the most recent file in
Context/Progress Updates/Q[N]-YYYY Weekly Wraps/ to compare trajectory (determine the correct quarterly folder from today's date: Q1 = Jan–Mar, Q2 = Apr–Jun, Q3 = Jul–Sep, Q4 = Oct–Dec)
- Slack — if Slack MCP tools are available, pull the week's biggest activity:
- Search DMs and key channels for messages involving your name, team, or active initiatives
- Look for: decisions in threads, action items, blockers, escalations, announcements
- Channels to prioritise: your key team channels and DMs from priority stakeholders (configure in GOALS.md or Context/Memory/)
- If Slack MCP tools are unavailable, note this and proceed — don't block the wrap
- Project tracker / Confluence — if MCP tools are available, check for status changes on key initiatives. If unavailable, skip.
Step 2: Internal Review (not shown to user)
Before writing the status update, do this analysis privately. All names, bias labels, scratchpad metrics, and detailed breakdowns stay in this step — they NEVER appear in the shareable portion (Step 3).
- Map completed work to goals — which goals moved, which didn't
- Identify the 4-6 most significant things that happened (decisions, deliverables, meetings, shifts)
- List every blocker and stalled item with owner and how long it's been stuck
- Pull specific names from meeting notes for action items (for internal sections only)
- Check against
Context/Memory/bias.md Section C — did any bias pattern show up this week? (for learnings only, never in shareable)
- Note product metrics: ticket volumes, SRR, CSAT, experiment results (these DO belong in shareable)
Step 3: Write the Weekly Status Update
Produce ONLY the short, shareable version directly. No long draft first. Hard limit: 250 words max for the shareable portion. This is sent to your manager — write it like a Slack message to them, not a report filed into a system.
Use this structure:
# Weekly Wrap CW[XX] — [DD–DD Mon YYYY]
**TL;DR:**
[2-3 sentences. Start with what happened, not a hype adjective. Keep problems for "On my mind". Positive but honest.]
## Update this week (selection)
- [Thing that moved, why it matters. Use accurate verbs (drafted/decided/kicked off, not shipped/approved).]
- [Combine related items. Include product metrics when available (ticket reduction %, CSAT, experiment results).]
- [Include one "strong take": an opinionated position on something you care about, not just status.]
(4-5 bullets max. ~25 words per bullet. Vary sentence structure, don't put an em-dash after every bold phrase.)
## On my mind the coming weeks
- [Strategic tension or question you're working through — not a to-do item]
- [Open question for the reader, if relevant — "keen to get your take on X"]
- [Risk or dependency you're watching]
(2-4 bullets. Reflective, not tactical. No specific dates or task names. Use "I" voice. Include asks when you need input.)
Voice rules
This reads like a message to your manager, not a status report.
- Conversational, direct — as if you're catching them up over coffee or in a Slack DM
- Address the reader when relevant: "per your suggestion last week", "keen to discuss in our 121", "would love your take on this"
- Show personality — excitement ("really promising"), concern ("worried about"), genuine curiosity ("wondering if")
- Include asks when you need input: "would appreciate your steer on X", "flagging for your awareness"
- At least one bullet should be a "strong take" — an opinionated position, not just facts: "Strong take: evals is the most important AI skill in 2025" or "I think the root cause is X, not Y"
- Use accurate verbs: "drafted", "kicked off", "exploring" — NEVER "shipped", "approved", "finalized" unless truly done
- Include real product metrics when available: ticket reduction %, SRR, CSAT, experiment hit rates
- Add context that explains WHY something matters, not just WHAT happened
Absolute rules for the shareable portion
- NEVER use individual names. No "shared with Sally", "aligned with Kerry", "Simone session". Use roles, teams, or unnamed references: "shared with design leads", "aligned with the team", "got positive signals from leadership". Names belong ONLY in the internal sections (Status Check, CW Focus).
- NEVER include internal system metrics — scratchpad completion rates, avoidance pattern counts, bounded time blocks, bias labels. These are your private productivity tools, not reader-facing content.
- NEVER include goal tags like "— Goal: AIRR" in the shareable portion. Save for internal Status Check.
- NEVER write a changelog. Don't list completed items. Tell the story of what moved and why it matters.
Exclude:
- Leadership philosophy or culture musings
- Long-term strategic hypotheticals
- Generic process improvements without a concrete owner
- Corporate jargon: "landed", "big calls", "key wins", "driving", "leveraging", "synergies", "key opportunity", "critical milestone", "significant progress", "major breakthrough", "major unlock", "emerged as a concrete integration opportunity"
- Salesy/hype openers in TL;DR: "Strong finish", "Great week", "Productive week", "Big progress". Just start with what happened.
- Em-dash overuse: don't put an em-dash after every bold phrase. Vary sentence structure. Use periods, colons, or just continue the sentence naturally.
- Implementation details the reader doesn't need: "reframed around X, dropped Y rationale, added Z context"
Good example (target this voice):
# Weekly Wrap CW21 — 19–23 May 2025
**TL;DR:**
Strategy playbacks went well with founders, positive signals on H2 direction. LLM eval goal moving with first vendor sessions done. Free user support showing early wins at 25% rollout.
## Update this week (selection)
- Strategy reviews done, positive feedback on H2 direction, no big concerns from leadership
- Had first sessions with two LLM eval vendors. Cost model and integration depth are the key tradeoffs. I regret not doing this earlier, it looks very promising
- All-hands presented H2 roadmap to 60+ attendees, good energy and alignment
- Free user support ramped to 25%, seeing 6K fewer tickets/month while CSAT holds stable
## On my mind the coming weeks
- AI Help integration needs to kick off cleanly while I'm away. Want to make sure handover is tight
- Ops restructure feels like it's being done to the team rather than with them. I think the root cause is centralised control, not collaboration structure. Keen to discuss
- LLM eval tool decision coming up. Leaning toward one vendor but want to validate cost model first
Bad example (what NOT to produce — every line violates a rule):
# Weekly Wrap CW12 — 16–20 Mar 2026
**TL;DR:**
Strong finish — all 4 top priorities shipped by Friday. CN AI proposal rewrite
broke a 3-week avoidance pattern with a bounded 90min block. Scratchpad
completion jumped from 17% to 52%.
[❌ Internal metrics. ❌ Avoidance/bias language. ❌ Changelog tone.]
## What happened
- Quarterly review first cut shared with design leads — synthesises key metric
trajectory, bypass impact, retention findings
[❌ Names 3 people. ❌ Implementation detail.]
- Initiative rewrite completed — reframed around core leverage, 3 weeks of
rolling finally broken with a bounded 90min block
[❌ Names nobody but explains the recipe. ❌ Internal avoidance tracking.]
- Operating model aligned — session (Mar 18) set ownership pairs, staged
lifecycle, 8-week cycles
[❌ Names person. ❌ Date in shareable.]
## On my mind the coming weeks
- I want to get the initiative proposal to leadership before the quarter ends
[❌ Names the manager. ❌ This is a task, not a tension.]
Step 4: Write the Full Wrap (internal detail, appended below the shareable update)
This section is internal — names, goal tags, and detailed breakdowns are OK here (this is where they belong, not in the shareable portion above). After the shareable update, append these sections for the saved wrap file:
## Status Check
Track the **team goals from your project tracker**, not personal objectives. These are the goals your team owns.
| Team Goal | Owner | Status | Notes |
|-----------|-------|--------|-------|
| [Goal name] | [Owner] | ✅ / ⚠️ / 🔴 / ⏸️ | [short note on progress or blocker] |
## CW[XX+1] Focus
**Must do:**
1. [item] — Goal: [which goal]
2. ...
**Should do:**
1. [item]
2. ...
**Watch:**
- [anything at risk]
## Learnings (saved to Memory)
- [Single sentence per learning. No elaboration. Pattern or insight in ~15 words max.]
- [Example: "Bounded time blocks break avoidance — 90min cap shipped the CN AI rewrite."]
- [3-5 learnings total]
Flag if any active quarterly objective has had zero activity two weeks running.
Step 5: Compound Learnings into Memory
Distill the week into durable knowledge for future sessions.
- Read
Context/Memory/learnings.md — avoid repeating what's already captured
- Extract 3-5 learnings as single sentences (~15 words each):
- Insight, pattern, or bias flag — no elaboration, no multi-sentence explanations
- Format: "[Pattern/insight] — [concrete example or evidence in same sentence]"
- If something keeps recurring across multiple wraps, promote it to a standing principle
- Save directly — append a new dated section to
Context/Memory/learnings.md (3–5 bullets max)
- Do NOT ask for confirmation — save and show what was added in the output
Compounding Principle
Each wrap should make the next one better:
- Reference prior learnings when assessing this week
- Promote repeated patterns into standing principles
- Prune stale learnings that no longer apply
- If
learnings.md grows past ~50 entries, suggest consolidating into higher-level principles
Step 6: Week Archival & Kanban Sync
-
Run a quick internal check (do NOT include in wrap output — this is housekeeping, not reader-facing):
- Count active task files (Tasks/ root, not Done/) missing
description field
- Check if
Context/Memory/active-tasks.md is older than 7 days
- If
Context/Memory/graph.yaml exists, capture node / edge counts and note stale or orphaned graph entries
-
Archive the week file:
- Move
Tasks/Week-YYYY-WNN.md to Tasks/Done/Week-YYYY-WNN.md
- This keeps the Tasks/ root clean — only active week files live there
-
Carry-forward tasks:
- Check the week file's "Deferred" / "Carry-forward" sections for uncompleted items
- These get picked up by
/plan-week when the next week is planned
- If any task has rolled 3+ weeks without progress, flag it explicitly in the wrap output
-
Kanban sync — Weekly Kanban.md has 4 columns: Backlog → This Week → In Progress → Done
- Move completed tasks (
status: d) to "Done" column, mark with [x] — remove them from their source column (In Progress, This Week, or Backlog)
- Move items that were "This Week" but didn't complete back to "Backlog" or "In Progress" as appropriate — remove from "This Week"
- Clear the "This Week" column (it gets repopulated by
/plan-week)
- Add the closed week file to "Done":
- [x] [[Week-YYYY-WNN]] — CW[XX] closed ([summary])
- Dedup check: before finishing, scan the entire board and remove any item that appears in more than one column (keep the most-advanced column: Done > In Progress > This Week > Backlog). Ensure all Done items are marked
[x].
-
Promotion check — for each completed task, check if it produced promotable outputs:
- PRDs, strategy docs, decision records → flag for
Context/Document Hub/
- Frameworks, playbooks, templates → flag for
Context/Knowledge/ or Context/Reference/
- Stakeholder insights → flag for
Context/Reference/stakeholders.md
- Report: "Promotable outputs: [list with suggested destinations]" — let the user confirm before promoting
-
Graph maintenance — if Context/Memory/graph.yaml exists:
- Halve the weight of any edge whose
last_seen is older than 4 weeks
- Prune edges with weight below 0.1 after decay
- Remove orphan nodes with no incident edges only if they do not have a
file: pointer
- Report: "Graph: N nodes, M edges. Decayed X, pruned Y."
-
If your workspace uses derived indexes and any are stale (>7 days), regenerate them or flag them for cleanup. Do not assume /rebuild-indexes exists.
Step 7: Save the Wrap
Determine the correct quarterly folder from today's date:
- Jan–Mar →
Context/Progress Updates/Q1-YYYY Weekly Wraps/
- Apr–Jun →
Context/Progress Updates/Q2-YYYY Weekly Wraps/
- Jul–Sep →
Context/Progress Updates/Q3-YYYY Weekly Wraps/
- Oct–Dec →
Context/Progress Updates/Q4-YYYY Weekly Wraps/
Save the full wrap (shareable update + internal detail sections) as CW[XX].md inside that folder.
If a file for this CW already exists, append the new wrap at the top (most recent first) rather than overwriting.
Confirm: "Saved to Context/Progress Updates/Q[N]-YYYY Weekly Wraps/CW[XX].md. Learnings appended to Memory."
Key Principle: No Mid-Flow Prompts
The entire wrap — shareable update, internal detail, learnings save, and file save — runs as one uninterrupted pass. The user reviews the complete output and provides corrections after, not during. This eliminates back-and-forth rounds.
If any external tool (Slack, Jira) is unavailable, note it and continue. Never block the wrap waiting for a tool.