Help us improve
Share bugs, ideas, or general feedback.
From buidl
Resumes an interrupted loop session from its last checkpoint. Reports session status, phase, cycle, worktree path, branch, and PR details, then continues execution.
npx claudepluginhub bc1plainview/buidl-opnet-pluginHow this command is triggered — by the user, by Claude, or both
Slash command
/buidl:buidl-resumeThis command is limited to the following tools:
The summary Claude sees in its command listing — used to decide when to auto-load this command
# The Loop — Resume Resume an interrupted loop session from its last checkpoint. Use this after context exhaustion, timeouts, or manual cancellation when you want to continue where you left off. ## Steps ### 1. Find State File Check for state files in order: 1. `.claude/loop/state.yaml` 2. `.claude/loop/state.local.md` (legacy fallback) If neither exists, report: "No loop session found. Start a new one with /buidl." ### 2. Parse State Read from the state file: - `session_name` — identifies the session directory - `status` — current status (done/failed/cancelled/timed_out are terminal...
/resumeResumes a named workflow from previous session, loading state files like workflow-state.json and HANDOFF_CONTEXT.md, validating git branch/commit/conflicts, restoring context, and showing phase progress with pending tasks.
/continueResumes prior session by scanning progress files and git state, summarizes work status with branch/phase details and next action, then loads context after confirmation.
/fire-loop-resumeResumes a Power Loop from Sabbath Rest snapshot, stopped state, or session interruption, restoring full context. Optional LOOP_ID; defaults to most recent.
/checkpointSaves current workflow state by identifying active workflows, updating progress, reconciling with git worktree, and outputting a markdown checkpoint summary with phase details and resume instructions.
/resumeResumes in-progress Forge execution after context reset or interruption by running forensic recovery, loading handoff context and state files, then continuing tasks.
Share bugs, ideas, or general feedback.
Resume an interrupted loop session from its last checkpoint. Use this after context exhaustion, timeouts, or manual cancellation when you want to continue where you left off.
Check for state files in order:
.claude/loop/state.yaml.claude/loop/state.local.md (legacy fallback)If neither exists, report: "No loop session found. Start a new one with /buidl."
Read from the state file:
session_name — identifies the session directorystatus — current status (done/failed/cancelled/timed_out are terminal; anything else is resumable)current_phase — where we left offcycle — current build-review cyclemax_cycles — cycle limitworktree_path — path to the git worktreeworktree_branch — branch nameproject_type — opnet or genericagent_status — which agents have completed (for build phase)tokens_used — tokens consumed so farpr_url — PR URL if one was createdworktree_path. If not:
git show-ref --verify refs/heads/<branch>git worktree add <path> <branch>.claude/loop/sessions/<name>/Look for .claude/loop/sessions/<name>/checkpoint.md. If it exists, read it for:
Display to the user:
Loop Resume
───────────
Session: [name]
Status: [status]
Phase: [current_phase]
Cycle: [cycle] / [max_cycles]
Worktree: [path]
Branch: [branch]
PR: [url or "not created yet"]
Tokens used: [tokens_used]
Checkpoint: [found/not found]
Resuming from: [phase description]
Update state: set status to the appropriate active status for the phase.
Use bash ${CLAUDE_PLUGIN_ROOT}/scripts/write-state.sh key=value for all state updates.
challenge → Re-read .claude/loop/sessions/<name>/challenge.md if it exists and continue the challenge Q&A. If challenge.md is complete, advance to specify.
specify → Check if spec files exist in sessions/<name>/spec/. If all three (requirements.md, design.md, tasks.md) exist, present them for approval. If incomplete, continue generating.
explore → Re-launch the explorer agents. Previous context.md is likely stale after a context reset.
build → This is the most common resume point. Read agent_status from state to determine which agents completed:
pending or in_progress: dispatch itdone: skip it (artifacts should exist)current_step to know where in the plan to resumereview → Re-launch the loop-reviewer with the current PR.
done/failed/cancelled/timed_out → These are terminal states. Report the final status and ask:
After resuming into the correct phase, follow the same phase logic as defined in the main /buidl command. The stop-hook will manage subsequent cycles automatically.