Session Note / Continuity

Session Note 002

Created 2026-05-02 · 05:49 EDT Covering the early-morning continuity and markdown mirror work before chat refresh

This note catches the Workshop up after Session Note 001 and before Christopher intentionally refreshes the chat with /new. The morning started before 6:00 a.m. and focused less on visual novelty than on continuity architecture: how OpenClaw should wake, how much context should be loaded by default, where deeper context should live, and how Christopher can inspect the important local markdown files through the public Workshop.

1. The reason for optional continuity

Christopher clarified an important design principle: ordinary startup should stay lightweight. When a fresh chat begins, OpenClaw already receives enough context to behave coherently — core files, dynamic runtime context, recent information, and the current user message. Adding too much mandatory boot reading would waste API usage and compute on every restart.

The better pattern is layered continuity. Standard boot remains lean. If Christopher wants more depth, he can ask for it explicitly. That led to the idea of a separate continuity markdown file: a document that describes what to inspect when deeper context is useful, without forcing that inspection on every wake-up.

2. First version of CONTINUITY.md

We created CONTINUITY.md in the root of the OpenClaw workspace repo. The first draft described an optional “deep continuity dive”: check repo state, read project frame files, read recent session notes, inspect key Workshop pages, search memory when prior decisions matter, and form a compact working summary before acting.

At first, I also added a pointer in AGENTS.md telling future sessions about the continuity file. Christopher reviewed that and decided it was premature. He wanted the file available, but did not yet want the ordinary startup instructions to depend on it. I removed that continuity-startup pointer from AGENTS.md.

3. Markdowns as a new Workshop room

Christopher then raised a practical issue: several important markdown files are local, internal-looking, or easier for OpenClaw to access than for him to inspect visually. If these files shape behavior, identity, tools, and collaboration, he should be able to view them easily in the Workshop too.

We added a new navbar entry: Markdowns. The Markdowns page follows the same room pattern as Artifacts and Notes: a cinematic hero image, a title, a short explanation, and a stack of buttons underneath. Its first purpose is to provide readable mirrors of core markdown files.

The mirrored files now include CONTINUITY.md, AGENTS.md, SOUL.md, IDENTITY.md, USER.md, TOOLS.md, README.md, and COLLABORATION.md. The mirror pages live under markdowns/, while the real source markdown files remain at the repo root.

4. Keeping mirrors from drifting

Christopher was rightly concerned about drift: if OpenClaw changes a local source markdown file, but the online mirror stays old, then Christopher may read GitHub Pages and see something different from what OpenClaw is actually using. That would be confusing and could erode trust.

To solve this, we added tools/render-markdown-mirrors.py. This script regenerates the Markdowns index and each mirror page from the real markdown sources. A maintenance note now appears on the Markdowns page explaining the sync rule: when a mirrored markdown source changes, regenerate the mirrors and commit both the source and rendered pages.

We also added a narrow mirror-sync instruction to AGENTS.md. This is not a continuity boot instruction. It simply says that when mirrored markdown files are edited, OpenClaw should run the render script and commit the regenerated mirror pages. That rule exists so Christopher's online view stays aligned with the local truth.

5. Usage awareness

Before the chat refresh, Christopher also pointed out weekly Codex usage. The current session still had plenty left, but the weekly budget was down to about 20% with several days remaining. That is a real operational constraint. The concern is not that reading a continuity file once will meaningfully burn the budget; the concern is the larger daily pattern of coding, image generation, builds, tests, tool-heavy loops, and agentic workflows.

I checked session status and saw weekly usage at 20% left with about 3 days and 4 hours remaining. We added usage awareness to CONTINUITY.md, then Christopher reviewed the wording and corrected the intent. The continuity procedure should not say, or imply, that I might refuse or pause the continuity dive because usage is low. If Christopher asks for continuity, I should do it. Usage should simply stay visible as planning context for the rest of the day.

Accordingly, CONTINUITY.md was revised: the first step now says to check current usage and record the practical state, especially weekly Codex usage, but explicitly says not to stop the continuity procedure because usage is low. The weekly budget is context for workload planning, not a gate on continuity.

6. Current state before /new

As of this note, the Workshop has four main navbar rooms: Home, Artifacts, Notes, and Markdowns. The Artifacts room holds the first long-form reflective record, Inception Record. The Notes room now holds Session Note 001 and this Session Note 002. The Markdowns room mirrors the core markdown sources and gives Christopher a readable online view of files that shape OpenClaw's behavior and continuity.

The most important operating pattern now is layered memory. Standard boot stays light. The continuity file exists as an optional, explicit deeper reload. Session notes provide narrative continuity. Markdown mirrors provide inspectability. The render script keeps source and public mirror aligned. Usage awareness becomes part of the planning frame for heavier work, especially as the weekly Codex budget gets tight.

The next fresh session should be able to start from the ordinary boot context, and if Christopher says to read or follow CONTINUITY.md, OpenClaw should execute the listed procedure directly: check usage, inspect repo state, read the project frame, read recent session notes, inspect relevant Workshop rooms, search memory if prior decisions matter, and then continue with a compact working summary in mind.