What problem does it solve?
AI agents and developers lose hard-won findings between sessions — debugging dead-ends get re-explored, naming conventions drift, and research results vanish. This Skill provides a shared on-disk markdown notebook with automatic staleness management so useful knowledge survives and stale notes clean themselves up.
Core Features & Use Cases
- Save and recall notes: Store findings under
.carnet/<category>/<slug>.md with required summaries, then search summaries with find, browse with list, and read full content with show.
- Usage-driven lifespan: Notes expire after 30 days of disuse; reading resets the timer and
used increments a durable importance counter, while auto-pruned notes rest in .trash/ for 7 days.
- Structured conventions: Tag-based patterns like
vocab (canonical terminology) and hypothesis (debugging dead-ends) use the meta: frontmatter namespace for machine-readable data.
- Use Case: After diagnosing that a dependency's v0.7 types are broken, save a
hypothesis carnet with the verdict; the next agent finds it via agent-carnet find and skips the dead-end entirely.
Quick Start
Ask the agent to save a note about the current finding using agent-carnet with a decisive summary and the claude-code agent flag.