groove-utilities-stats

name	groove-utilities-stats
description	Quantified compound-loop dashboard: lessons captured vs graduated, adherence streaks, and task velocity. Use to see whether the workflow is actually compounding.
license	MIT
allowed-tools	Read Glob Grep Bash(git:*) Bash(beans:*) Bash(gh:*) Bash(linear:*) Bash(jq:*) Bash(python3:*) Bash(chmod:*) AskUserQuestion
metadata	{"author":"andreadellacorte","bash":true}

groove-utilities-stats

Use $ARGUMENTS to specify the period: week (default), month, or all.

Bash fast-path

This skill has a bash script at scripts/stats.sh. Run it directly for faster, model-free execution:

bash .agents/skills/groove-utilities-stats/scripts/stats.sh week

(Pass week, month, or all to match $ARGUMENTS.) If the script exits 0, report its stdout and stop — do not continue with the steps below. If it exits non-zero, continue with the markdown steps (exit 3 means it printed the local metrics but the linear/github task counts still need to be computed below; exit 1 means no .groove/).

Outcome

A read-only dashboard is printed to the conversation quantifying compound-loop health for the period: the compound funnel (lessons captured vs graduated), adherence (daily-log coverage, streak, rollup freshness, ratings), and velocity (task open/closed, commits). No files are written. This answers "is the workflow compounding / am I keeping the rhythm" — for qualitative patterns, it points to /groove-utilities-memory-retrospective.

Acceptance Criteria

• Compound funnel shown: lessons captured (total + in period), graduated (total + in period), graduation rate
• Adherence shown: daily-log coverage over the last memory.review_days business days, current streak, weekly/monthly rollup freshness, ratings count + average
• Velocity shown: task open/closed counts (if a task backend is configured) and commit count in the period
• Output goes to the conversation only; nothing is written
• Absent data sources are skipped with sensible "n/a"/"NA" placeholders rather than errors

Steps

(Model fallback — the bash fast-path above does this deterministically. Reproduce the same output format.)

1. Memory path is always .groove/memory/. If .groove/ is absent, stop with an install hint.

2. Read .groove/index.md frontmatter: tasks.storage (backend; legacy key tasks.backend) and memory.review_days (default 5).

3. Determine date range from $ARGUMENTS: week = last 7 days, month = last 30 days, all = everything. Call the start <since>.

4. Compound funnel:

   • Captured: count ## YYYY-MM-DD dated entries across .groove/memory/learned/*.md (exclude signals.md) — total, and those with date ≥ <since>.
   • Graduated: count [graduated YYYY-MM-DD] markers across the same files — total, and those ≥ <since>.
   • Graduation rate = graduated_total / captured_total as a percentage (n/a if none captured).

5. Adherence:

   • Daily-log coverage: of the last review_days business days (Mon–Fri, ending today), how many have .groove/memory/daily/YYYY-MM-DD.md.
   • Streak: consecutive most-recent business days (back from today) that have a daily log; stop at the first gap.
   • Rollup freshness: weekly is fresh if a .groove/memory/weekly/<ISO-year>-W<ISO-week>.md exists for the current or previous week, else stale; monthly likewise for .groove/memory/monthly/YYYY-MM.md.
   • Ratings: parse .groove/memory/learned/signals.md table rows (| YYYY-MM-DD | N/5 | note |) with date ≥ <since>; report count and average (and a ▁▂▄▆█ sparkline if ≥ 3).

6. Velocity:

   • Commits: git log --oneline --since=<since> count.
   • Tasks, by tasks.storage:
     • beans: open = beans list --json --no-status completed --no-status scrapped | jq length; closed = --status completed + --status scrapped; best-effort open mistakes/promises = count of non-terminal children under the "Mistakes"/"Promises" epics.
     • linear: via the linear CLI/MCP — assigned issues grouped into open vs done.
     • github: gh issue list --assignee @me --state all grouped into open vs closed.
     • none: omit task counts (show commits only).

7. Print the dashboard in the Output format below.

Output format

## Groove Stats — <period> (<start> to <today>)

### 🔁 Compound funnel
Lessons captured: <N> (<n> this <period>)   Graduated: <G> (<g> this <period>)
Graduation rate: <pct>

### 📿 Adherence
Daily logs: <x>/<review_days> business days   Streak: <s> days
Rollups: weekly <fresh|stale>, monthly <fresh|stale>
Ratings: <c> in period | avg <a>/5  <sparkline>

### 🚀 Velocity
Tasks: <open> open / <closed> closed   Commits: <k> in period
Open: <m> mistakes, <p> promises

_For qualitative patterns and learnings, run `/groove-utilities-memory-retrospective`._

Constraints

• Read only — never write to any file (no memory, no AGENTS.md, no config).
• Skip any absent data source with a one-line n/a/NA rather than erroring; parse tables and dates defensively.
• Paths are hardcoded under .groove/memory/ (not configurable), consistent with the rest of groove.
• This skill complements /groove-utilities-memory-retrospective (qualitative reflection) — do not duplicate its narrative; cross-link it.