Carry project memory across coding agents with OpenWolf
Use OpenWolf to keep local project memory, handoff packets, project maps, and token usage records available across Claude Code, Codex, and OpenCode sessions.
npx skills add agentskillexchange/skills --skill carry-project-memory-across-coding-agents-with-openwolf
Use OpenWolf when an operator needs a coding agent to resume work with local, reviewable project context instead of rereading the same files and replaying the same decisions. The workflow is concrete: initialize OpenWolf in a repository, save checkpoints and known fixes into the local .wolf folder, build or refresh a project map, query focused symbols or files before large reads, and export/import handoff packets when work moves between Claude Code, Codex, and OpenCode.
What this skill actually does
Invoke this instead of relying on a product’s normal chat history when the user is switching agent hosts, recovering after compaction, handing off between sessions, or trying to reduce repeated context reads in an active codebase. It is strongest when the operator wants the memory to stay local, inspectable, and separate from any provider account.
This is not a generic memory product listing. The scope boundary is repo-local coding-agent continuity: checkpoints, handoffs, project maps, token/activity records, and local file-backed context. Do not stretch it into general personal knowledge management, cloud memory, or broad RAG.
Inputs and prerequisites: OpenWolf CLI; a target code repository; supported Claude Code, Codex, or OpenCode workflow; optional dashboard inspection.
Setup notes: Install the openwolf package from npm or follow the upstream getting-started guide, then run OpenWolf commands from the target repository. Use openwolf find and openwolf map for focused repo lookup, and checkpoint/handoff commands for session continuity.
Source and verification boundary: use https://github.com/cytostack/openwolf and https://openwolf.com as the canonical sources. Verify behavior against upstream command docs before relying on a generated handoff or token report.
Framework fit: publish as Multi-Framework because upstream explicitly targets Claude Code, Codex, and OpenCode and the approved workflow is cross-agent project memory rather than a single-host extension.