recover-claude-code

{
  "name": "recover-claude-code",
  "version": "1.2.3",
  "description": "Bring back Claude Code sessions, workflows and subagents that a crash interrupted, with the context they already had instead of restarting them from scratch. Use after a terminal, Ghostty or machine crash, a force-quit, or a jetsam/out-of-memory kill took sessions down, and whenever a previous recovery cold-started fresh agents and lost what they had already worked out. Separates the genuinely dead sessions from the still-running ones using the live-session registry at ~/.claude/sessions (a pid-keyed record carrying sessionId, cwd, peer name and status), because the two probes that look correct both fail: ps output never contains the session id, and the scratchpad directory is named by a per-process id that stops matching once a session has been resumed. Reopens each dead session in its own Ghostty tab at its original working directory under its original session id (never --fork-session, which orphans every workflow journal it is trying to reach), handing over a brief as a positional prompt so the session wakes up knowing it crashed rather than auto-submitting \"Continue from where you left off\". Restores interrupted work by promoting a dead subagent's sidechain transcript into a resumable session so it keeps the files it read and the findings it closed, rather than replaying its prompt into a blank context; measured, a promoted transcript named its item, branch and a closed finding from memory. Finds a run's persisted script by run id across every project directory, because Claude Code files the script under the project directory of the live working directory and the journal under the original one, so any session working in a worktree splits its own state in two and a snapshot-only lookup reports no script for precisely the runs that need recovering. Offers a journal splice for a long completed prefix that resumeFromRunId would otherwise re-run (the cache miss is sticky, so the interrupted