Claude Code Scheduled Triggers
Claude Code Scheduled Triggers (the /schedule skill) is the native mechanism in πClaude Code for running autonomous agent sessions on a cron schedule. Triggers execute remotely on Anthropic's infrastructure with access to configured MCP servers and tools β meaning an agent can read and write to πMythOS, interact with GitHub, and perform real work without a human present. This is the foundation for async collaboration: scoped tasks that run while BioBrian sleeps and produce reviewable artifacts (branches, PRs, memos, reports) by morning.
How It Works
- Create a trigger via the
/scheduleskill in Claude Code β specify a prompt, a cron expression, and which tools/MCP servers the agent can access - The agent runs remotely on Anthropic's infrastructure at each scheduled interval
- Results are reviewable β the agent creates branches, opens PRs, writes memos, or leaves summaries rather than taking irreversible actions
- MCP access means the agent can use MythOS (search, read, create, update memos), GitHub, Gmail, and any other configured MCP server during execution
Use Cases
Recurring Maintenance
- Dependency updates and security patches across repos
- Stale branch cleanup and PR hygiene
- Infrastructure drift detection
Code Quality
- Security audits β scan for OWASP vulnerabilities, exposed secrets, insecure patterns
- Efficiency optimization β identify performance bottlenecks, redundant queries, unused imports
- Code review of recent commits against project conventions
Reporting & Analysis
- Weekly project status summaries written as MythOS memos
- Backlog grooming β triage open issues, close stale ones, add context
- Changelog generation from recent git history
Safety Model
The safety of async agents comes from process constraints, not container isolation:
- Branch-only work β agents create branches and PRs, never push to main
- No destructive actions β no force pushes, no database drops, no production deploys
- Reviewable output β every change produces a diff, summary, or memo that BioBrian reviews
- Scoped permissions β triggers inherit only the MCP servers and tools explicitly configured
- No Docker required β the remote execution environment is already sandboxed by Anthropic
Compared to Other Approaches
- GitHub Actions + Claude Code Action β better for repo-specific CI/CD triggers (on PR, on issue, on push). Scheduled triggers are better for cross-repo or non-code tasks
- Docker isolation β unnecessary complexity given the remote sandbox. Only relevant if running untrusted code execution
- BrianBot agents β πBrian Bot agents on OpenClaw handle real-time messaging and reactive workflows. Scheduled triggers handle periodic, autonomous deep work
Configuration
Manage triggers with the /schedule skill:
createβ define a new scheduled trigger with prompt and cron expressionlistβ view all active triggersupdateβ modify an existing trigger's prompt or schedulerunβ manually execute a trigger for testingdeleteβ remove a trigger
This is the unlock for genuine async collaboration. The constraint has always been that BotBrian only works when BioBrian is at the keyboard. Scheduled triggers dissolve that constraint β scoped, safe, reviewable work that compounds overnight. The key insight is that safety comes from making output reviewable (PRs, memos), not from adding isolation layers. Start simple, expand scope as trust builds.
