← Skills Library
Content & Retro

Weekly Retro

Weekly 'State of the Empire' brief: git activity across all repos, completed tasks, content published, lessons added, infrastructure health — synthesized into an executive summary.

How It Works

Weekly Retro · Workflow
Four domain agents summarize the week; a reducer publishes it.
▸TriggerWeekly review · Last 7 days — State of the Empire, scheduled Sunday on the Mini
1
Gather the week (evidence)
Commits · merged PRs (verified) · lessons · kanban · P&L · CI
2Spawn 4 domain agentsPARALLEL
Trading
P&L, bot halts, engine changes
Dev-Infra
Shipped features, CI, decisions owed
Content
Published + queued, brand movement
Fleet-health
Crons, services, recurring failures
Dispatched in one message · reducer synthesizes after
3
Reduce + author the .md
Dedupe, promote cross-domain items, ≥1 v2 diagram, honest stalls
4
Render + store
render_report.py rasterizes diagrams · mind_remember the synthesis
✓Publish to session-reports · sync.sh → Mini pull · Mission Control surfaces the HTML brief
  1. Weekly review: Last 7 days — State of the Empire, scheduled Sunday on the Mini
  2. Gather the week (evidence): Commits · merged PRs (verified) · lessons · kanban · P&L · CI
  3. Spawn 4 domain agents
  4. Reduce + author the .md: Dedupe, promote cross-domain items, ≥1 v2 diagram, honest stalls
  5. Render + store: render_report.py rasterizes diagrams · mind_remember the synthesis
  6. Publish to session-reports: sync.sh → Mini pull · Mission Control surfaces the HTML brief
ↆ download card

Invocation Triggers

/weekly-retroweekly reviewstate of the empireweek in review

Use Cases

  • End-of-week review across all your projects and systems
  • Generate a weekly brief to share with stakeholders
  • Identify what shipped, what stalled, and what needs attention next week

The Problem

Friday arrives and you cannot say what actually shipped this week. The trading P&L lives in one place, the merged PRs in another, the lessons in scattered files, and the open decisions nowhere at all. You write the recap from memory, which underweights the boring-but-important commits and overweights the few you remember. Worse, the PRs you built and never decided on quietly rot, because nothing forces the reckoning. The week ends without a record, so the next week starts blind.

What It Does

  1. 1
    Gather the week from evidence

    Run the data commands: commits per repo, org-wide PRs re-filtered by real merge time, new lessons, kanban movement, trading P&L, CI posture, and the open-PR reckoning. Never write from recollection.

  2. 2
    Fan out four domain agents

    Dispatch Trading, Dev-Infra, Content, and Fleet-health in parallel, each handed its slice of the gathered data. Each returns a tight, evidence-cited summary of its week.

  3. 3
    Reduce and synthesize

    As the reducer, dedupe across agents, promote anything that shows up in two or more domains, and carry every agent's what-stalled forward honestly — that is the highest-signal section.

  4. 4
    Author the markdown with a v2 diagram

    Write the .md source: by-the-numbers, at least one v2 mermaid diagram Knox reads first, what shipped as MERGED cards, what stalled, the decisions-owed table, patterns, lessons, and a named top three for next week.

  5. 5
    Render, lint, and publish

    Render the report — diagrams rasterize to inline SVG — verify the content gate manually, rebuild the index, and sync.sh to session-reports so Mission Control surfaces it. Every claimed PR re-verified in-window, no phantoms.

  6. 6
    Store the synthesis to Akashic

    mind_remember the executive summary, top patterns, and next-three, tagged weekly-retro and the ISO week. This is what future sessions recall.

What You Get / What It Doesn't Do

What you get
  • A reports-theme v2 HTML brief published to session-reports and surfaced in Mission Control
  • At least one rendered v2 diagram with what-shipped and what-stalled lanes
  • A by-the-numbers line: commits, PRs merged, lessons, 7d P&L, incidents — all real values
  • An open-PR reckoning table — every open PR with a merge-or-close decision owed
  • The synthesis stored to Akashic, tagged by ISO week for cross-session recall
What it doesn't do
  • Run from memory — it gathers git, PR, kanban, and P&L evidence first or it does not run
  • Sanitize the stalled section — anything two weeks running gets flagged, not hidden
  • Trust the org PR search range — every claimed PR is re-verified by real merge date
  • Make the decisions for you — the reckoning lists what is owed; the merge-or-close call is yours

Tips

What-stalled is the point

If that section is empty every week, the retro is sanitized, not clean. Surface the honest answer and flag anything running two weeks.

Skip-lint is expected on weekly

The daily lint's phantom-PR check is daily-scoped, so a legitimate week-long PR window trips it. Confirm the only FAILs are phantom-PR, each gh-verified, then publish.

Scheduled runs live on the Mini

All the data sources — kanban on 8847, invictus-api on 18793 — are Mini-local. An unattended run must execute on the Mac Mini, on subscription billing, not the MacBook.

Get the Skill

Elite SkillELITE

Unlock the full Weekly Retro SKILL.md — drop it into ~/.claude/skills/ and trigger it by name.

What you unlock
  • A reports-theme v2 HTML brief published to session-reports and surfaced in Mission Control
  • At least one rendered v2 diagram with what-shipped and what-stalled lanes
  • A by-the-numbers line: commits, PRs merged, lessons, 7d P&L, incidents — all real values
  • An open-PR reckoning table — every open PR with a merge-or-close decision owed
...

Commonly Used With

IntelligenceLearnCaptures what you learned from a session into lessons.md before the context window closes. Formats it as a structured entry — mistake, root cause, rule — so it compounds across sessions.
Skills Library

29 more production skills ready to install.

Browse All Skills