From fluent
Atomically updates 6 Fluent learner databases (learner-profile, progress, mistakes, mastery, spaced-repetition, session-log) at practice session end via Python script and JSON payload. Persists errors, review results, new vocab, and metadata.
npx claudepluginhub m98/fluent --plugin fluentThis skill uses the workspace's default tool permissions.
Every practice skill ends with a DB update. Instead of hand-editing 6 JSON files (error-prone, racy, easy to desync), pipe one JSON report to `update-db.py`. The script runs pre-write backups, validates the payload, applies all changes atomically via `.tmp + fsync + rename`, and rebuilds the spaced-repetition queue.
Runs interactive vocabulary drill sessions using spaced repetition, flashcard prompts across recognition, production, and cloze modes, with per-answer scoring and batched DB updates via Python/Bash. Triggered by /vocab.
Reviews session to propose lean updates to existing skills, capturing debugging patterns, data issues, queries, doc references, and workflow improvements. Invoke at session end.
Analyzes session history via ccrecall.db or in-context to extract learnings from corrections, discoveries, and failures, then proposes persistent skill updates. Invoke /reflect post-session.
Share bugs, ideas, or general feedback.
Every practice skill ends with a DB update. Instead of hand-editing 6 JSON files (error-prone, racy, easy to desync), pipe one JSON report to update-db.py. The script runs pre-write backups, validates the payload, applies all changes atomically via .tmp + fsync + rename, and rebuilds the spaced-repetition queue.
Load this skill whenever the tutor:
total_sessions, current_streak_days, or total_study_minutes.Skip this skill for read-only operations (use the progress skill or read-db.py directly) and during session setup (use setup skill instead — update-db.py is for session deltas, not bootstrap).
Run from the repo root:
python3 .claude/hooks/update-db.py <<'EOF'
{ ...payload... }
EOF
Exit codes: 0 success, 1 validation error, 2 I/O error.
Required fields
session_id — string, convention session-NNN. Use computed.next_session_id from read-db.py.date — YYYY-MM-DD.Optional fields — omit to skip. Full canonical example (copy-paste this and fill in):
${CLAUDE_PLUGIN_ROOT:-${CLAUDE_PROJECT_DIR:-.}}/.claude/references/db-updater-payload.example.json
Key blocks the example covers: skill_scores, errors[], new_vocabulary[], review_results[], topics_covered, breakthroughs, focus_next_session, session_notes, achievements_earned, milestones.
errors[] — one entry per distinct mistake this session. Collapse duplicates (same pattern_id) before sending; frequency is bumped by the script.new_vocabulary[] — items the learner met for the first time. Fill every field; incomplete entries yield incomplete spaced-repetition records.review_results[] — items already in the queue that were reviewed. The script runs SM-2 on each. See the sm2-calculator skill. Mapping: quality = floor(score / 2).skill_scores[].correct counts correct exercises, not a percentage. Accuracy is derived.confidence in learner-profile.skills is 0–100 integer; accuracy in progress-db is 0.0–1.0 float. The script handles the conversion.Always call read-db.py at session start to get current state + next_session_id. Don't read each JSON file separately:
python3 .claude/hooks/read-db.py
Returns all 6 databases plus computed fields (due_reviews_count, next_session_id, streak_active, days_since_last_session).
python3 .claude/hooks/update-db.py <<'EOF'
{
"session_id": "session-012",
"date": "2026-04-24",
"duration_minutes": 12,
"command_used": "/review",
"skills_practiced": ["vocabulary", "grammar"],
"skill_scores": {
"vocabulary": { "exercises": 3, "correct": 3, "time_minutes": 7 },
"grammar": { "exercises": 2, "correct": 1, "time_minutes": 5 }
},
"review_results": [
{ "item_id": "vocab_huis", "quality": 5 },
{ "item_id": "vocab_deur", "quality": 4 },
{ "item_id": "vocab_raam", "quality": 5 },
{ "item_id": "grammar_omdat_word_order", "quality": 2 },
{ "item_id": "grammar_past_tense", "quality": 4 }
],
"errors": [
{
"pattern_id": "grammar_omdat_word_order",
"category": "grammar",
"your_answer": "omdat ik ben moe",
"correct_answer": "omdat ik moe ben",
"context": "subordinate clause word order",
"severity": "critical"
}
],
"focus_next_session": ["Drill 'omdat' word order"]
}
EOF
python3 .claude/hooks/update-db.py <<'EOF'
{
"session_id": "session-013",
"date": "2026-04-25",
"command_used": "/vocab",
"skills_practiced": ["vocabulary"],
"new_vocabulary": [
{
"item_id": "vocab_keuken",
"item_type": "vocabulary",
"content": "de keuken",
"answer": "the kitchen",
"category": "household_rooms",
"difficulty": "A1",
"initial_quality": 4,
"priority": "medium"
}
]
}
EOF
spaced-repetition.review_queue. It's regenerated from scratch on every run.session_id replaces. Sending the same ID twice overwrites the first call. Useful for corrections, dangerous if unintentional..backups/pre-update-<session_id>/ before any change. Check there to roll back.Six interdependent JSON files must agree: a new session-log entry, a bumped total_sessions, updated SM-2 params, new mistake patterns, recalculated accuracy, refreshed streak. Hand-editing causes silent desync — streak says 7 days but session-log has 6 entries, mastery says 4 stars but accuracy says 45%. The script is the single source of truth.