Description
Summary I run a small OpenClaw setup (6 agents) that was built conversationally over a few months rather than engineered. It works, but it has grown messy: memory files are bloating, API spend is higher than it should be, and one agent reads an inbound email account and acts on what it finds. I'm looking for someone to review the entire setup and prepare a report on what they find. This is a time-boxed review, not an ongoing engagement, though I expect follow-on remediation work if the review is good What I need reviewed: - Agent and skill structure : how the agents, skills, a are laid out, where there's duplication, what should be consolidated or modularized - Cost : where the spend is actually going by agent and by model, and specifically which jobs are burning frontier-model calls on work that belongs in a deterministic script or a local model - State and memory : the Markdown-file memory is growing without bound and sessions are overloading. I want an opinion on the right storage model and a migration path - Permissions : what each agent can actually reach, and where scope should be tightened Deliverable: A written findings document, 2 to 4 pages, that a non-technical person could follow. Ranked by impact, with rough effort estimates against each recommendation. No code changes required in this engagement. What I'll provide: Read-only SSH access to the staging host (Mac Mini, Apple Silicon), the config directory, and 30 days of API usage exports. To apply, please answer: - An OpenClaw or comparable agent fleet you've worked on: how many agents, what broke, what you did about it - The largest LLM cost reduction you've delivered - In two or three sentences, how you'd secure an agent that reads and acts on inbound email