Post-close analysis that reads how the close actually went — pace, review cycles, reassignments, reopens, workload balance. Produces a Slack-ready digest organized around questions a controller would actually ask, not raw statistics.
Before You Begin
- Numeric MCP connected.
- Numeric Toolkit installed.
- At least one closed period (or an in-progress period — will be labeled "progress-to-date").
- (Optional) Slack MCP for posting the digest to a channel.
How It Works
- Ask Claude for a retro. Example prompts:
- "How did the close go?"
- "Close retro for March 2026."
- "Compare this close to last quarter."
- "What took longest in Feb?"
- Claude asks which period and pulls tasks and events
- Claude computes defined metrics: turnaround, handoff rate, reopens, coverage gaps, workload concentration.
- Claude generates a plain-language digest andffers to post to Slack.
What You Get
A Slack-formatted digest with 7 sections:
- Quick numbers — KPI snapshot (completion %, turnaround, handoff rate, reopens, coverage gaps, churn)
- How's the pace? — first-touch latency, turnaround distribution
- Is the review process working? — handoff rate, reopens, rework cycle time, self-review
- Is work spread evenly? — concentration, most-loaded person, fastest/slowest, dormant tasks
- Setup gaps — unassigned tasks, missing due dates, high reassignment
- What went well / What to watch — auto-derived from the data
- Suggested actions — 1–3 concrete next steps
Optional dashboard: Ask Claude to generate a dashboard for you with scorecard tiles, pace chart, late-tasks table, review-friction table, assignee leaderboard.
Troubleshooting
| Issue |
What to Check |
"No events returned" |
Falls back to status-only metrics. Efficiency metrics skipped. |
Slow on large workspaces |
Events API is the bottleneck. The skill caps batches at 15 task keys; 6–10 batches is enough for reliable medians. |
Negative turnaround values |
Data quirk where assignment was recorded after submission. The script filters these automatically. |
No prior period available |
Period-over-period comparison is skipped. |
Digest feels generic |
Ask the skill for specific questions from references/questions.md — the question bank is customizable. |
FAQ
- Can I run it mid-close? Yes. Labels everything as "progress to date" and replaces cycle time with "Day X of Y."
- Medians with <10 tasks? Works but noisy. Stick to raw counts for small periods.
- Can I compare retros? Yes — say "compare this close to last quarter" and the skill pulls both periods.
Download & Install
Download the full Numeric Toolkit (15 community skills, including this one):
🔗 numeric-mcp-toolkit.zip
Install via Claude Code CLI:
/plugin marketplace add geoff-lew/numeric-toolkit
/plugin install numeric-mcp-toolkit
Need help? Reach out to support@numeric.io.