Open-source universal operating harness for AI agents using a portable rules core plus work-type profiles for coding, research and other workflows.
AgentSmith is PromptPartner’s universal operating harness for AI agents. Instead of supplying a model, it assembles a lean rules core plus work-type profiles into instruction files such as CLAUDE.md, AGENTS.md and GEMINI.md. The current repository documents 10 profiles covering software development, DevOps/setup, marketing/outreach, document creation, data work, administration, deep research, creative design, security audit and autonomous loops. 1
The harness also includes verification, handoff, research/feedback, secret-scan and leak-gate scripts; optional skills/plugins; self-update and doctor flows; PowerShell and bash setup; and cross-tool output for agents including Claude Code, Codex, Gemini, OpenClaw and Cursor. These are documented project capabilities, not AiToolMap hands-on findings. 1
AgentSmith is free open-source software under the MIT License. The current repository exposes an explicit LICENSE file and describes commercial reuse/adaptation as permitted subject to retaining the notice. AiToolMap found no separate paid AgentSmith software tier. 1
The harness itself does not eliminate any subscription or API costs associated with the AI coding assistants or models a user chooses to run underneath it.
The repository documents cautious and trusted permission presets, verification-before-completion gates, secret scanning, leak gates, dirty-checkout protection during self-update and circuit-breakers around especially destructive commands. The trusted mode deliberately permits broad tool use and the project itself warns about deletion, exfiltration and remote-push risk on sensitive machines. 1
Specialist directories independently surface the repository, MIT license and current adoption/activity signals; SkillsLLM also reports an automated vulnerability/prompt-injection scan. AiToolMap treats that scan as limited automated context, not an independent security certification. 2 3
The canonical 50-source panel found no exact independent benchmark, mainstream review corpus or eligible numeric rating input. No hands-on testing was performed.
AgentSmith is best suited to users who already rely on one or more agentic coding/work assistants and want a reusable operating layer for rules, verification, safety, handoffs and profile-specific quality gates. It is especially useful where the same operating discipline needs to follow work across projects or model vendors. 1
It is less suitable for users looking for a standalone AI model, a hosted no-code agent builder, enterprise support commitments or independently proven productivity gains.
explicit MIT license; broad work-type profile coverage; support for multiple instruction-file ecosystems; bash and PowerShell installers; verification and handoff primitives; safety presets; secret/leak tooling; transparent source; meaningful early GitHub adoption signals. 1
young repository history; no exact independent benchmark or mainstream review corpus found; effectiveness of the workflow philosophy is not independently quantified; powerful trusted-mode automation can materially increase local-machine risk if used without appropriate boundaries. 1
A preview from the separate AiToolMap video library. These videos complement the written review; they are not part of the rating.
Agentsmith
Official demo embedded on the Agentsmith product site, showing its prompt-CMS workflow for authoring, testing, syncing and deploying AI-agent prompts. Original YouTube metadata was not separately retrievable, so the displayed title is an editorial label.
WATCH ON YOUTUBE ↗Products matched by shared tasks, audience, workflow, product type and capabilities.