| name | observation-rollup |
| description | Compresses observation logs into weekly summaries. Rolls completed weeks once they are 3+ days past week-end AND have ≥7 days of data. Say "roll up observations" or "compress old logs" (or run /founder-os:observation-rollup). Archives the source JSONL files. Idempotent and recoverable - safe to run anytime.
|
| why | Prevents the brain/observations/ folder from accumulating thousands of raw JSONL lines by compressing old weeks into readable summaries you can actually review. |
| enhance | Set FOUNDER_OS_OBSERVATIONS=1 in your shell to generate JSONL files in the first place - the rollup has nothing to process if observation logging is off. |
| summary | Roll up weekly observation logs. |
| allowed-tools | ["Read","Bash"] |
| mcp_requirements | [] |
Observation Rollup
Runs on: local-exec - runs a local script; on a cloud surface I read the results, I do not run it.
Compress old observation logs into weekly summaries. Runs scripts/observation-rollup.py and reports the result.
When to use
Run this skill when the SessionStart brief surfaces a nudge about unrolled JSONL files older than 10 days. Also safe to run anytime - idempotent runs produce no changes on already-rolled weeks, and the default mode archives source files rather than deleting them so a rollup is always recoverable.
Triggered by: "roll up observations", "compress old logs", "compact observations", /founder-os:observation-rollup.
Procedure
- Verify
scripts/observation-rollup.py exists at the repo root.
- Run:
python scripts/observation-rollup.py
- Report the output: what rolled, what was skipped, any errors. The output for each rolled week tells the founder whether sources were archived or deleted.
- If rollups were written, note the count of summary files now in
brain/observations/_rollups/ and remind the founder that source JSONLs are in brain/observations/_archived/<week>/.
What the script does
- Walks
brain/observations/*.jsonl. Filenames must be YYYY-MM-DD.jsonl.
- Groups files by ISO week (
YYYY-Wnn).
- For each week with >= 7 days of data AND today >= 3 days past the week-end:
- Aggregates: total observations, tool counts, top 5 tools, skill invocations, unique session IDs.
- Writes
brain/observations/_rollups/YYYY-Wnn.md.
- Default: MOVES source JSONL files to
brain/observations/_archived/YYYY-Wnn/ once the rollup is verified written. Recoverable.
- With
--delete-sources: deletes source JSONL files instead. Irreversible. Use only if archive disk usage is a concern and the rollup contents are trusted.
- Weeks with < 7 days of data or that ended < 3 days ago are skipped.
- Weeks already rolled are skipped (idempotent).
Cleaning up archived sources
brain/observations/_archived/ accumulates over time. There is no automatic retention sweep. Delete subfolders by hand when no longer needed, or run python scripts/observation-rollup.py --delete-sources on a future rollup if you trust the past rollups and want to skip the archive step going forward.
Requirements
- Python stdlib only. No pip installs, no API keys, no external services.
FOUNDER_OS_OBSERVATIONS=1 must have been set to produce any JSONL files.- If
brain/observations/ is empty, the script exits cleanly with no changes.
