What problem does this solve?
Grok Build (grok, xai-org/grok-build) is xAI's open-source terminal coding agent. Spool currently prepares and reads sessions from Claude Code, Codex CLI, Gemini CLI, OpenCode, and Pi — but not Grok Build.
Heavy Grok Build users accumulate a large local session history (in my case ~3 GB under ~/.grok/sessions/) that Spool cannot index, organize, or read. Since Spool's local preparation already covers five agents, Grok Build is the most notable gap: several community tools (agent-hop, claudescope, seshport) have already added parsers for this format, which suggests real demand.
What would the solution look like?
A new parser in packages/core/src/parsers/grok.ts, following the existing per-provider parser pattern. The on-disk format is simple and stable:
- Sessions live at
~/.grok/sessions/<url-encoded-cwd>/<session-uuid>/
chat_history.jsonl — one JSON record per message: {type, content}summary.json — session metadata: agent_name, chat_format_version, created_at / updated_at / last_active_at, current_model_id, num_messages, session_summary, reasoning_effort, etc.updates.jsonl — {method, params, timestamp} records carrying per-turn timestamps and token usage
Scope-wise, Link-only visibility (like Gemini CLI / OpenCode / Pi today) seems like the right starting point; native resume could follow later since Grok Build supports grok --resume <session-uuid>.
Prior art that may help: seshport PR #1 is a worked example of adding Grok Build support to a session tool; agent-hop and claudescope also ship parsers for this format.
I'm happy to contribute a PR for this if the direction sounds right.
What problem does this solve?
Grok Build (
grok, xai-org/grok-build) is xAI's open-source terminal coding agent. Spool currently prepares and reads sessions from Claude Code, Codex CLI, Gemini CLI, OpenCode, and Pi — but not Grok Build.Heavy Grok Build users accumulate a large local session history (in my case ~3 GB under
~/.grok/sessions/) that Spool cannot index, organize, or read. Since Spool's local preparation already covers five agents, Grok Build is the most notable gap: several community tools (agent-hop, claudescope, seshport) have already added parsers for this format, which suggests real demand.What would the solution look like?
A new parser in
packages/core/src/parsers/grok.ts, following the existing per-provider parser pattern. The on-disk format is simple and stable:~/.grok/sessions/<url-encoded-cwd>/<session-uuid>/chat_history.jsonl— one JSON record per message:{type, content}summary.json— session metadata:agent_name,chat_format_version,created_at/updated_at/last_active_at,current_model_id,num_messages,session_summary,reasoning_effort, etc.updates.jsonl—{method, params, timestamp}records carrying per-turn timestamps and token usageScope-wise, Link-only visibility (like Gemini CLI / OpenCode / Pi today) seems like the right starting point; native resume could follow later since Grok Build supports
grok --resume <session-uuid>.Prior art that may help: seshport PR #1 is a worked example of adding Grok Build support to a session tool; agent-hop and claudescope also ship parsers for this format.
I'm happy to contribute a PR for this if the direction sounds right.