context-digest

name	context-digest
description	Prepopulate agent context with high-salience oc-memory cells by auto-generating a Memory Digest section in MEMORY.md or CLAUDE.md. Use when setting up oc-memory for the first time, wiring memory into a new agent workspace, or when the agent should have top memories pre-loaded at session start without running manual searches.

Context Digest

Keeps a ## Memory Digest section in your agent's context file fresh so every new session starts with high-salience memories already loaded — no manual search needed.

Works with OpenClaw (MEMORY.md), Claude Code (CLAUDE.md), and any agent platform that auto-injects a project file at session start.

How It Works

1. A cron job (or manual run) executes gen-context-digest.sh every 2-3 hours
2. The script queries mem scene <name> for each configured scene, filters by salience ≥ 0.8, and rewrites a bounded block in the target file
3. Your agent platform auto-injects the file at session start → digest is in context before the first user message

Target File Auto-Detection

The script picks the target file in this order:

1. Explicit argument: bash gen-context-digest.sh /path/to/file.md
2. MEMORY.md in $WORKSPACE (OpenClaw convention)
3. CLAUDE.md in $WORKSPACE (Claude Code convention)
4. Falls back to $WORKSPACE/MEMORY.md (creates on first cron run)

Setup

1. Copy the script

cp skills/context-digest/scripts/gen-context-digest.sh ~/bin/gen-context-digest.sh
chmod +x ~/bin/gen-context-digest.sh

2. Test it

WORKSPACE=/path/to/your/workspace bash ~/bin/gen-context-digest.sh

Verify the ## Memory Digest block appears in your target file.

3. Add HTML markers to your target file

Add this block where you want the digest to appear:

<!-- ARCHY_DIGEST_START -->
<!-- ARCHY_DIGEST_END -->

The script will replace between the markers on next run. If no markers exist, it appends to the end of the file.

4. Install cron

crontab -e

Add (adjust paths):

0 */3 * * * export PATH="$HOME/bin:/usr/local/bin:/usr/bin:/bin:$PATH"; WORKSPACE=/path/to/workspace bash ~/bin/gen-context-digest.sh >> /tmp/context-digest.log 2>&1

Platform-Specific Notes

OpenClaw

OpenClaw auto-injects MEMORY.md at session start. Add the markers to MEMORY.md and the digest is available before the first user message.

For AGENTS.md integration:

## First-Message Recall

**Memory Digest is pre-loaded** in `## Memory Digest` at the bottom of `MEMORY.md`.
Use it immediately. Run `mem search "<topic>"` only for specific topics not covered by the digest.

Claude Code

Claude Code auto-injects CLAUDE.md at session start. Add the markers to CLAUDE.md:

# My Project

... your existing CLAUDE.md content ...

<!-- ARCHY_DIGEST_START -->
<!-- ARCHY_DIGEST_END -->

The digest will be in context for every claude session without any MCP tool calls.

For projects using both MEMORY.md and CLAUDE.md, point the script at whichever file your agent reads first:

WORKSPACE=/path/to/project bash gen-context-digest.sh /path/to/project/CLAUDE.md

Customizing Scenes

The script queries these scenes by default: lessons, config, infrastructure, projects, foreman, commodore.

Edit the scene list in gen-context-digest.sh to match your memory schema:

LESSONS=$("$MEM_CMD" scene "my-lessons" ...)
INFRA=$(  "$MEM_CMD" scene "my-infra" ...)

Run mem scenes to list all scenes in your store.

Force Refresh

WORKSPACE=/path/to/workspace bash ~/bin/gen-context-digest.sh