| name | work-log-ingestion |
| description | Turns Draft Editor work-log entries marked ingest=true into durable knowledge: append to a digest markdown file and re-run meta-layer graph ingest. Use for weekly sweeps, habit capture follow-up, or when the user says to process the work log / ingest links from the calendar. |
Work log → knowledge ingestion
Source of truth
- File:
data/master_calendar.json → array work_log
- UI: Draft Editor → Calendar tab → Work log (or POST /api/work-log with
ingest: true)
Entries may include text, status (started / finished / note), attachments (URLs or paths), ingest (boolean; default true on create).
When to use
- User: “Process the work log” / “Ingest this week’s work log” / “Sweep work_log for ingestion”
- Standing habit: weekly (e.g. Monday) run a sweep after dictation habit
- After a burst of links_or_paths captures
Workflow
1. Sweep digest (idempotent)
From agent-lab root:
uv run python scripts/sweep-work-log-ingest.py
This script:
- Reads
work_log from data/master_calendar.json
- Skips entries with
ingest: false
- Skips entries already present in
knowledge/work_log_digest.md (matched by <!-- wl-id:{id} --> markers)
- Appends new sections under
## headings so ingest-meta-layer-knowledge can chunk them
2. Graph ingest
After the sweep, load chunks into Neo4j (same pipeline as other knowledge markdown):
uv run --project framework/deer-flow/backend python scripts/ingest-meta-layer-knowledge.py --file knowledge/work_log_digest.md
If the digest is new or large, a full knowledge run also picks it up when listed in the script’s default file set (see ingest-meta-layer-knowledge.py).
3. Optional agent steps (no script)
- Summarize high-value
finished entries into JAUmemory remember (decisions, patterns).
- For URLs in attachments, fetch/summarize and use add-to-meta-layer-graph / user’s graph tools when appropriate.
Rules
- Do not re-append the same work-log
id—the sweep script guards with HTML comments.
- Secrets: Never paste tokens or private URLs into the digest; redact if needed.
- Prefer sweep script + ingest for bulk; use manual graph add for one-off gold nuggets.
Related
- master-calendar skill — API and calendar semantics
- ingest-meta-layer-knowledge.py — Neo4j chunks for RAG
