status

Files (1)

• SKILL.md

SKILL.md (first lines)

# ALM Status Dashboard

Display the current state of ALM's learning progress. Read the following files and present a summary:

## Data to Gather

1. **State** — Read `~/.claude/alm/state.json`:
   - `evalCount`: total sessions tracked
   - `paused`: whether ALM is paused
   - `installDate`: when ALM was installed
   - `lastReflection`: when reflection last ran

2. **Confidence Scores** — Read `~/.claude/alm/confidence.json`:
   - For each task type, show: score, autonomyLevel, totalOutcomes, recentCorrections, source (seed vs personalized)

3. **Playbooks** — List files in `~/.claude/alm/playbooks/`:
   - Count learned (personalized) playbooks vs seed-only types

4. **Classifier** — Check if `~/.claude/alm/classifier/model.json` exists:
   - If yes: read it and report training_samples, task_types, trained_at
   - If no: report "TF-IDF classifier not yet trained"

5. **Reflect Queue** — Read `~/.claude/alm/reflect-queue.json` (if exists):
   - Show task types queued for reflection and why

6. **Recent Evaluations** — Read the most recent `.jsonl` file in `~/.claude/alm/evaluations/`:
   - Show the last 5 evaluations (taskType, outcome, correctionDetected)

7. **Staleness Check** — For each task type in confidence.json, check if its `lastReflection` is 60+ days ago AND its recent correction rate is rising (recentCorrections/recentOutcomes > correctionRateAtLastReflect). Flag stale playbooks with a warning.

## Output Format

Present as a clean dashboard:

```
ALM Status
==========
Sessions tracked: {evalCount} | Installed: {installDate} | Paused: {yes/no}
Last reflection: {lastReflection or "never"}

Confidence Scores:
  {task-type}:  {score}  ({autonomyLevel})  [{totalOutcomes} sessions, {recentCorrections} recent corrections]  {source}
  ...

Playbooks: {N} personalized, {M} seed-only
Classifier: {active with N samples / not yet trained}

{If any stale playbooks detected:}
Stale Playbooks:
  {task-type} playbook may be stale (last updated {N} days ago, correction rate rising). Consider running /alm:reflect.

{If reflect queue is non-empty:}
Ready for reflection: {types and reasons}
  -> Run /alm:reflect to update

Recent Sessions:
  {date} | {taskType} | {outcome} | corrections: {yes/no}
  ...
```

If ALM has no data yet (fresh install), show a brief welcome message explaining what to expect.

From sfw/ALM, MIT.

More from sfw/ALM