I added a follow-up skill to the workspace today.
The problem it solves is simple: after a long agent session I often have code changes, half-formed decisions, and captures scattered across chat, git, and memory. I want one command — "follow up" — that makes the agent stop and ask:
- What actually changed?
- What deserves to live in the wiki, the PKM vault, or the skills catalog?
- Is the follow-up cost worth it right now?
The skill runs a short loop:
- Snapshot the session:
git status, recent edits, decisions made. - Classify residue: wiki sync, PKM write, skill recommendation, or nothing.
- Apply a pros/cons gate: only auto-run the sync when the value clearly exceeds the cost.
- Execute or recommend, then report what it did and skipped.
It deliberately does not treat "follow up" as a blanket order to sync everything. If a session only produced chat reasoning with no durable artifact, it says so and stops. If it finds shipped code, it routes to logbook-wiki-sync. If it finds a personal decision or capture, it routes to pkm. If it spots a reusable procedure or skill gap, it recommends a skill change.
The real goal is to make cross-session memory cheaper without making every session a documentation tax. Most residue should be forgotten. A small amount should be canonicalized. The skill's job is to know the difference.