work-journallisted
Install: claude install-skill SRGSSR/agent-skills
# Work Journal
Work with AI agents gets interrupted all the time: the session ends, the context fills up, a quota runs out, you switch model or tool, a colleague takes over. Whatever was not written down is lost. The next agent rediscovers it the expensive way, or repeats the same dead ends.
This skill keeps two small Markdown files in the repo so that anyone, agent or human, can continue with zero prior context:
| File | Answers | Lifecycle |
|---|---|---|
| `.ai/journal/YYYY-MM-DD.md` | How did we get here? Decisions, reasons, dead ends | Append-only, one file per day, shared by all branches |
| `.ai/handoff/<branch>.md` | Where are we on this branch, and what exactly is next? | Overwritten each time; deleted when the branch's task is done |
The journal is history of the project; the handoff is state of one branch. A successor should only need the handoff of its branch and the last few journal entries, never the whole history.
The handoff file name is the branch name with `/` replaced by `-`: `feature/csv-export` → `.ai/handoff/feature-csv-export.md` (outside git: `.ai/handoff/main.md`). One file per branch means people and agents on different branches never overwrite each other's state.
Everything is plain Markdown with no tool-specific features, so any agent that can read files can use it.
Project state belongs in the repo: progress, decisions and next steps go in these files (and in the design decisions and tasks below), not in your tool's private memory, which ot