AgentSkillsCN

digest

进阶

SKILL.md
--- frontmatter
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/:

FileWhat to extract
opportunities.mdOptimization ideas - "could improve", friction points, repeated manual work, better approaches mentioned
loose-ends.mdOpen items - TODOs not done, questions unanswered, "need to" without resolution
research-backlog.mdFuture directions - "interesting", "worth exploring", novel techniques
blog-grist.mdSignificant work completed, problems solved, patterns discovered
things-learned.mdPreferences, environment facts, workflow learnings

Instructions

1. Find unprocessed transcripts

bash
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):

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

2. Extract and analyze

For each session with new content:

bash
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:

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

Or pipe multiple lines:

bash
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:

bash
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.