executor-loop

name	executor-loop
description	Run repeated manual executor cycles bead-by-bead until the ready queue is exhausted or a blocker requires user input. Use when the user wants sequential autonomous progress without swarm coordination.

Executor Loop

Run repeated manual executor cycles bead-by-bead until the queue is exhausted or a blocker requires user input.

For epic-scoped multi-agent work with coordinator-owned bead state, prefer swarm-epic.

Steps

1. If the current repo is not initialized for Beads, stop and tell the user to run the template bootstrap script or at minimum bd init --prefix <prefix> plus the repo scaffolding steps.
2. Determine the first bead:
   • if the user supplied a bead id in the current request, start there
   • if the user supplied freeform selector text, treat it as a selector or hint for the first bead
   • otherwise inspect bd ready --json and choose the best ready bead autonomously
3. Run one full executor cycle for that bead by invoking every step in order:
   • beads-claim
   • writing-plans
   • implementation
   • systematic-debugging if blocked
   • build-and-test after implementation; read .codex/skills/build-and-test/SKILL.md and follow it
   • verification-before-completion or requesting-code-review
   • beads-close
4. After a successful close and local commit, inspect bd ready --json again and choose the next best ready bead using the same preference order.
5. Repeat until one of these stop conditions is reached:
   • no ready bead remains
   • a blocker requires user input
   • build, test, or verification cannot pass
   • manual intervention is required
6. When stopping on a blocker:
   • do not auto-resume
   • summarize the current bead, the blocker, and what input or fix is needed
   • wait for the user to continue in normal chat
7. When stopping because no ready work remains, summarize the completed beads and any follow-up beads created during the loop.

Hard Rules

• Treat each bead as its own logical executor cycle.
• Do not hold multiple claimed beads at once.
• Never continue past a blocker without user input.
• This is a sequential path. Do not market it as true multi-agent swarming.