Why logbooks?
The idea is to equip skills with state that outlives the run — somewhere their work piles up across sessions, across agents, across runs.
Most skills have a natural shape for their work. Make that shape explicit, and the skill stops starting from scratch — every run builds on the last.
What structure looks like
Three patterns where rows do what prose can't.
What this enables
Deep research and review
Build multi-phase work from structured intermediate results.
Draft and staging workflows
Shape work before committing it to Jira, docs, or reports.
Background agent supervision
Inspect progress without reading full transcripts.
Cross-agent reuse
Let each pass leave behind work the next pass can use.
Controlled execution
Turn selected entries into tickets, reports, boards, or new agent runs.
Flexible by design
A logbook can live in a CSV, spreadsheet, JSONL file, SQLite database, or a purpose-built tool. What matters is not the backend. What matters is that the work is structured enough to query, update, and carry forward.
Skill examples
Five published skills that exercise logbook principles.
$ /plugin marketplace add agentlogbooks/logbooksHotspot-first code review. ≤5 high-signal findings per run.
/plugin install deep-code-review@logbooks
ideation
Per-topic SQLite logbook. Operators + playbooks across sessions.
/plugin install ideation@logbooks
mutation-testing
Surviving mutants are real bugs your suite would miss.
/plugin install mutation-testing@logbooks
skill-creator eval workspace
JSON-file workspace for iterating on skills. evals.json + history.json + per-run grading. (external)
git clone anthropics/skills