From scheduled-execution
Three-tier scheduled execution model for Claude Code (cloud Routines, local persistent Desktop tasks, session-scoped /loop). Triggers on "run this every day", "schedule X", "automate the audit", "make this recurring", or any periodic-execution need. Built from reality-check Spec ST-001 framing + the architecture brief's three-tier model.
How this skill is triggered — by the user, by Claude, or both
Slash command
/scheduled-execution:scheduled-executionThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Three tiers, picked by cadence and trust requirements:
Three tiers, picked by cadence and trust requirements:
| Tier | Mechanism | Use when… |
|---|---|---|
| Cloud | Claude Code Routines (April 2026+) | Daily / hourly recurring work; runs on Anthropic infrastructure; no local machine needed |
| Local persistent | Desktop tasks (local Claude Code) | Daily / hourly recurring work; needs filesystem or local-only resource access |
| Session-scoped | /loop slash command in a session | Short-lived polling within an active session ("check the build every 5 minutes") |
Reality-check (the daily Orryx Group audit at D:\Orryx\reality-check\) is the canonical example of tier 2 / Local Persistent. The architecture brief's Wave 4 reality-check plugin formalises this pattern.
Best when:
Example use: daily audit of public-facing endpoints; daily Slack summary; weekly competitor analysis.
How to set up: Use the /schedule skill (Claude Code plugin) or Anthropic Console Routines panel.
Cost: per-run API cost. Tracks in usage dashboard.
Best when:
Example use: reality-check daily audit (scans D:\ project dirs); local backup verification; nightly local lint pass.
How to set up:
# Windows Task Scheduler — example for reality-check
$action = New-ScheduledTaskAction -Execute "claude" -Argument "code /reality-check"
$trigger = New-ScheduledTaskTrigger -Daily -At 06:00
Register-ScheduledTask -TaskName "Orryx-RealityCheck" -Action $action -Trigger $trigger
For macOS/Linux: launchd or cron respectively.
Cost: free except the API cost per run.
/loop)Best when:
Example use: "every 5 minutes check the deploy status until it's green or failed", "poll the PR for review comments", "wait for the test suite to complete then summarise".
How to use:
/loop 5m /deploy-status
Slash command runs every 5 minutes within the current session. /loop without an interval lets the model self-pace.
Cost: per-invocation API cost. Session ends → loop ends.
Is this a one-time task? → Just do it.
Does it need to run when my machine is off? → Tier 1 (Cloud Routines)
Does it need local filesystem / local secrets? → Tier 2 (Desktop tasks)
Is this just polling within a session? → Tier 3 (/loop)
D:\Orryx\reality-check\ runs daily at 0913 (per the audit's Phase 1 finding). It:
baselines/<repo>.json (fingerprint-keyed)reports/daily-reality-check.md (rolling summary)reports/archive/<run-id>.mdTier 2 (Desktop tasks) is correct because:
Once delivery/reality-check (Wave 4 plugin) ships, this should consume the audit's _entity-map.json for per-entity rollups.
| Cadence | Use cases |
|---|---|
| Daily | Reality-check; security scan; dependency audit; backlog grooming |
| Hourly | Health checks; cost-spike detection; lint of in-flight branches |
| Weekly | Sprint planning prep (per methodology/spec-driven-delivery); cross-repo drift audit |
| Per-PR | Code review; cross-model verification; quality gates |
/loop in unattended session. Tier 3 is for active sessions. For unattended, use Tier 2.delivery/reality-check (Wave 4) — the canonical Tier 2 exampledelivery/credential-audit (Wave 4) — could run at any tierdelivery/security-audit (Wave 4) — typically daily Tier 2D:\Orryx\reality-check\ — local reference implementationCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub alexmclaren/orryx-knowledge --plugin scheduled-execution