Placeholder repo for exploring an idea (and tooling) with Claude.
One layer of the agentic tooling stack has received comparatively little attention: the journal / logbook / documentation layer.
In traditional development, a junior dev might end a session by writing up what was achieved, what decisions were made, what's blocked, and where the project stands. That artifact is mundane but load-bearing — it's how the next person (or the same person next week) resumes without re-deriving context.
In agentic development, this matters even more. Agents lose context between runs, hand work off to other agents, and make decisions that a human supervisor needs to audit after the fact. Right now, most of the tooling around this is improvised:
- I scaffold folders like
planning/agent-logs,planning/on-hold,planning/project-mgmt,planning/decisions - I complement them with slash commands and subagents (e.g. a handover mechanism for passing logs between agents)
- I ask the agent to generate docs — sometimes reference docs for me ("here's what we set up and how to maintain it"), sometimes public-facing docs
A folder scaffold gets you part of the way, but it isn't an elegant long-term solution.
A dedicated journal / documentation layer for the agent — not general vectorised agentic memory, but a structured, mutually-legible record that both the agent and the human supervisor can read and write comfortably.
Imagine:
- The human checks into the agent's "wiki / logbook" to review yesterday's runs
- They can see where the agent got stuck, what it decided, and why
- A kanban-style view surfaces what's in progress, blocked, or done
- The agent reads the same surface to reorient at the start of a new session
This would sit alongside (not replace) task management, orchestration, and memory — as the documentation layer of the AI project-management picture.
Placeholder. Exploring tooling and shape. Nothing to install yet.