takurot

Agent Scheduler — Git skill for Claude Code

Git community

A subscription-aware coding agent scheduler.

How to install Agent Scheduler

This entry records only its repository, not the path inside it, so there is no exact command to give. Open takurot/agent-scheduler and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Scheduler does

A subscription-aware coding agent scheduler. Routes GitHub Issues to Claude Code and Codex within flat-rate quotas, featuring Git worktree isolation, multi-stage workflows, and automated TDD gates.

Alternatives in Git

README

Subscription Agent Scheduler (`subsched`)

`subsched` is a deterministic, subscription-aware coding agent scheduler that turns GitHub Issues into a durable task queue and routes tasks to coding agents (Claude Code, OpenAI Codex) with zero metered API fallback.

[!NOTE] **Package & Command Mapping**:

  • PyPI package: agent-scheduler (pip install agent-scheduler)
  • Python import: subsched (import subsched)
  • CLI command: subsched (primary) or agent-scheduler (alias)

Key Features

  • 🔒 Subscription-Only Execution: Strictly operates within flat-rate subscription quotas (Claude Pro/Team, ChatGPT Plus/Team). Prevents accidental pay-as-you-go API key charges with fail-closed safety.
  • 🌳 Worktree Isolation: Creates isolated Git worktrees (subsched/issue-) for each task to prevent workspace collision.
  • Reactive Multi-Agent Failover: Classifies rate-limit results returned by workers and preserves cooldown/reset state for failover. Proactive provider-capacity monitoring is not yet wired.
  • 🛡️ Automated TDD & Quality Gates: Enforces test-driven development, running repository verification (ruff, mypy, pytest with $\ge 80%$ coverage, pip-audit) before pull requests.
  • 🔀 Merge Conflict Protection: Resolves the repository default branch, fetches its latest origin state, rebases onto that remote-tracking ref, and cleanly aborts on conflict before escalating to NEEDS_HUMAN.
  • 📊 Observability & Metrics: Tracks autonomous completion rates, task success metrics, structured JSON Lines event logs, and markdown run reports.

⚠️ Security Notice: Native Container Boundary

`--allow-native` execution runs the Claude Code / Codex CLI with permission checks bypassed (`bypassPermissions`), so the agent can run Bash commands, edit files, and read files without per-command confirmation. This is required for unattended execution — in non-interactive mode there is no human to confirm anything,