Notre avis

Extrait des informations exploitables à partir de journaux de transcription et les organise dans des fichiers artefacts structurés.

Points forts

Catégorise automatiquement les insights en plusieurs fichiers (opportunités, tâches en suspens, etc.)
Intègre des références aux sessions sources pour une traçabilité complète
Interface en ligne de commande simple avec des commandes dédiées (list, extract, note, mark)
Gère les horodatages et le formatage des notes de manière automatique

Limites

Nécessite un format de transcription spécifique (sessions avec ID)
Le marquage manuel des sessions traitées peut être oublié
Ne traite que les transcriptions déjà enregistrées, pas le streaming en direct

Quand l'utiliser

Après une session de conversation pour capturer les idées clés, les tâches à faire et les apprentissages.

Quand l'éviter

Lorsque les transcriptions ne sont pas disponibles au format attendu ou que vous avez besoin d'une prise de notes en temps réel.

Exemples

Process latest transcript

Run the digest skill on the latest transcript session. First list unprocessed sessions, then extract insights from the newest one, and add notable entries to opportunities and loose ends files.

Extract and note specific session

Use digest to process session abc123: extract its content, add any research ideas to research-backlog.md, and mark the session as processed.

Bulk extract all new transcripts

Run digest list to see unprocessed sessions, then extract each one and add findings to the appropriate artifact files. Report a summary of what was added.

name: digest description: Extract insights from transcripts into actionable artifact files.

/digest

Process raw transcript logs and extract insights directly into artifact files.

Output Files

All in notes/digest/:

| File | What to extract | |------|-----------------| | opportunities.md | Optimization ideas - "could improve", friction points, repeated manual work, better approaches mentioned | | loose-ends.md | Open items - TODOs not done, questions unanswered, "need to" without resolution | | research-backlog.md | Future directions - "interesting", "worth exploring", novel techniques | | blog-grist.md | Significant work completed, problems solved, patterns discovered | | things-learned.md | Preferences, environment facts, workflow learnings |

Instructions

1. Find unprocessed transcripts

uv run --project ~/.claude/skills/digest digest list

If output is "Nothing new to process", stop here.

Use --path <dir> to specify a different base directory (defaults to cwd):

uv run --project ~/.claude/skills/digest digest --path /other/project list

2. Extract and analyze

For each session with new content:

uv run --project ~/.claude/skills/digest digest extract <session_id>

This outputs cleaned message text (user/assistant only, no metadata).

What to look for (include source ref [session_id:line] for each item):

Opportunities (opportunities.md):

"could improve", "should optimize", "better way"
Friction, repeated steps, workarounds
Failed approaches that revealed a gap

Loose ends (loose-ends.md):

TODOs not completed in the session
Questions raised but not answered
"Need to" / "should" without resolution

Research backlog (research-backlog.md):

"Interesting", "worth exploring"
Techniques mentioned but not tried
Links/references saved for later

Blog grist (blog-grist.md):

Significant work completed
Problems solved in interesting ways
Patterns or insights that emerged

Memories (things-learned.md):

User preferences ("I prefer...", "don't like...")
Environment facts (paths, configs, tools)
Workflow patterns to remember

3. Add notes

Use the note command to add entries. It handles timestamp and formatting automatically:

uv run --project ~/.claude/skills/digest digest note notes/digest/opportunities.md --project "project-name" "- Insight here [session_id:line]"

Or pipe multiple lines:

echo "- First insight [abc123:42]
- Second insight [abc123:55]" | uv run --project ~/.claude/skills/digest digest note notes/digest/opportunities.md --project "myproject"

The CLI prepends entries with a timestamp from date '+%Y-%m-%d %H:%M'.

4. Mark processed

After extracting insights from a session:

uv run --project ~/.claude/skills/digest digest mark <session_id>

5. Report

Brief summary:

Sessions processed: N (X new lines)
Items added: Y opportunities, Z loose ends, etc.

Keep it short. The artifacts speak for themselves.

Extraction d'insights depuis transcriptions

Recommandé pour