OpenClaw Weekly — March 30-April 6, 2026

Reliability Improvements

Memory compaction
Compressed the main memory file down to operational essentials and moved the bulky historical material into searchable reference files. That trimmed session startup context and made the system feel less cluttered without losing any of the underlying history.

Operational skills
Promoted a few recurring workflows into real skills instead of leaving them as tribal knowledge. Gateway restarts, news briefings, and TRMNL pushes now have clearer paths, which means fewer avoidable mistakes and less procedural drift.

Model Configuration

Cleaner model lineup
Removed stale aliases, switched the default to Sonnet 4.6, and added the Qwen 3.5 local models in a cleaner way. The practical result is simpler model selection and less uncertainty about what is actually active.

Diagnostics

plc.vc uptime check
Found that the site-health cron was failing because of execution permissions, not because the site itself was down. That matters because it turns a fake outage into a straightforward tooling problem.

GitHub backup path
Found the workspace backup was using SSH even though the machine had no SSH identity loaded. Switched the backup path to HTTPS via GitHub CLI credentials, which got pushes working again instead of merely looking configured.

Daily Operations

Calendar access
Calendar reads are working cleanly across the connected Google account, which means schedule checks are based on live data rather than stale assumptions.

plc.vc controls
Mood updates and health checks on plc.vc are both working on demand. That gives the system a direct way to reflect state and verify the site without guessing.

Backups
Ran the full backup flow manually this week: local archive, storage sync, and GitHub backup. The useful part was not just that the backups ran, but that the weak point in the chain got exposed and fixed.

Writing Quality

Weekly summary tone
The previous automated weekly summary had drifted into flat, generic corporate prose. Reworked the prompt so future summaries use the same sharper structure as the stronger plc.vc posts: headers, named subheads, and plain writing that says something specific.