Routines

Routines are periodic maintenance tasks that keep the project healthy. They audit dependencies, clean up documentation drift, validate project status, and scan for vulnerabilities — all on a defined schedule.

Routines are managed exclusively by the Tech Lead. Code agents don’t run them. This is by design — maintenance tasks require broad project context and cross-cutting awareness that only the orchestrating agent has.

How Routines Work

Each routine has three components:

Component	What it is
Frequency	How often it should run (e.g., every 5 days, every 7 days, every 14 days)
Procedure	A concrete, step-by-step process the Tech Lead follows
Dashboard entry	A row in agent/Core/routines.md tracking when it last ran and when it’s next due

Routines are opt-in. The routine-check sensor flags overdue routines at session start, but you decide when to run them. The agent never runs a routine autonomously — it waits for your approval.

The Routine Lifecycle

1. Detection

   At the start of every session, the routine-check sensor parses the dashboard at agent/Core/routines.md. It compares each routine’s last-ran date against its frequency and flags any that are overdue.

   Agent: OVERDUE: dependency-audit (last ran 12 days ago, due every 7)

2. Approval

   You tell the Tech Lead to run the routine. You can be specific or broad:

   You: Run the dependency audit
   You: Run all overdue routines

3. Execution

   The Tech Lead follows the routine’s concrete procedure step-by-step. Each routine has a detailed checklist — the agent doesn’t improvise. For example, the dependency-audit routine walks through scanning go.mod, checking for CVEs, flagging unmaintained packages, and documenting findings.

4. Logging

   Results are appended to station/Logs/RoutineLog.md with a timestamp, the routine name, and a summary of findings.

5. Dashboard update

   The Tech Lead updates the last_ran timestamp in agent/Core/routines.md. The next-due date is automatically calculated from the frequency.

Complete Routine Reference

All 8 routines in the Bonsai catalog:

Routine	Frequency	What it does
backlog-hygiene	7 days	Review Playbook/Backlog.md — flag stale items, escalate P0s, cross-reference with Status and Roadmap
dependency-audit	7 days	Scan dependencies for known CVEs and unmaintained packages
doc-freshness-check	7 days	Detect documentation drift — find docs that are stale relative to recent code changes
infra-drift-check	7 days	Compare declared IaC state against actual cloud resources
vulnerability-scan	7 days	SAST scan, secrets scan, and dependency audit — flag new, resolved, and persistent vulnerabilities
status-hygiene	5 days	Keep Playbook/Status.md clean — archive old Done items, validate Pending items
memory-consolidation	5 days	Bridge Claude Code auto-memory into agent/Core/memory.md — validate facts, clean redundant entries
roadmap-accuracy	14 days	Ensure Playbook/Roadmap.md reflects what’s actually built and what’s actually next

Note

All routines are currently compatible with the tech-lead agent type only. Other agent types don’t have routines by default — the Tech Lead is the only agent with the broad project context needed for maintenance tasks.

The Routine Dashboard

The dashboard lives at agent/Core/routines.md inside the Tech Lead’s workspace. It tracks each routine’s status in a structured format:

## Routines

| Routine | Frequency | Last Ran | Next Due |
|---------|-----------|----------|----------|
| backlog-hygiene | 7 days | 2025-01-15 | 2025-01-22 |
| dependency-audit | 7 days | 2025-01-10 | 2025-01-17 |
| doc-freshness-check | 7 days | 2025-01-14 | 2025-01-21 |
...

The routine-check sensor reads this dashboard at session start. If any routine’s next-due date is in the past, it flags it as overdue in the session context.

Auto-Managed Sensor

The routine-check sensor is automatically managed by Bonsai:

• Installed when you add any routine to an agent
• Removed when you remove the last routine from an agent

You never need to manually install or configure it. If an agent has routines, the sensor is there. If it doesn’t, the sensor is gone.

Routines and Scaffolding

Routines depend on scaffolding files to do their work:

Routine	Scaffolding dependency
backlog-hygiene	Reads and maintains Playbook/Backlog.md
status-hygiene	Reads and maintains Playbook/Status.md
roadmap-accuracy	Reads and maintains Playbook/Roadmap.md
doc-freshness-check	Scans all project documentation
dependency-audit	Logs results to Logs/RoutineLog.md
vulnerability-scan	Logs results to Logs/RoutineLog.md
memory-consolidation	Reads and maintains agent/Core/memory.md
infra-drift-check	Logs results to Logs/RoutineLog.md

This is why scaffolding modules like Playbook and Logs are required — routines rely on them to function.

Running Routines

You:    Hi, get started
Lead:   [startup flags: "OVERDUE: dependency-audit (last ran 12 days ago, due every 7)"]
You:    Run the dependency audit
Lead:   [follows procedure -> logs to RoutineLog.md -> updates dashboard]

Useful prompts for working with routines:

• “Are any routines overdue?”
• “Run all overdue routines”
• “Run the vulnerability scan”
• “What did the last dependency audit find?”

Each routine has a concrete, step-by-step procedure — the agent doesn’t improvise. This makes routine execution predictable and auditable.

Creating Custom Routines

You can create your own routines by adding markdown files with YAML frontmatter (including a frequency field) to agent/Routines/. See Creating Custom Routines for the full guide.