Vault Loop

Open-source prompt and skill for turning an Obsidian-style vault, project wiki, or second-brain folder into a maintained knowledge system.

Vault Loop helps an AI agent design recurring update lanes, state files, runbooks, review gates, and scheduler prompts. It is intentionally generic: it does not include private chat sources, local paths, account IDs, or platform-specific secrets.

For AI Agents

If a user gives you this repository and asks you to "set up my knowledge-base automation" or "turn this into an open-source prompt", do this:

1. Read PROMPT.md first. Treat it as the canonical behavior contract.
2. Read skill/vault-loop/SKILL.md if the user wants an Agent Skills compatible version.
3. Use templates/ only when producing concrete artifacts for a user's vault.
4. Ask for source scope and privacy boundaries before scanning files, chats, drives, browser data, or cloud services.
5. Create a small working loop before proposing many automations.
6. Never advance a state cursor unless the target window was actually covered and validation passed.
7. Run python scripts/validate_project.py before claiming the prompt pack is ready.

The output expected from you is not a long essay. The expected output is a working bundle:

• one or more lane prompts;
• a runbook per recurring lane;
• a state file with last_success_at, overlap, pending queue, and known blockers;
• a deterministic validation checklist or script where possible;
• clear active / paused / proposed status for each lane.

What It Builds

Vault Loop can be adapted to maintain:

• personal knowledge bases;
• team project wikis;
• research notebooks;
• meeting-note systems;
• source-monitoring workflows;
• daily reflection and status lanes;
• local health checks and backup routines.

It does not assume a specific scheduler. The same prompt pattern can be used with Codex heartbeat automations, cron, Windows Task Scheduler, macOS launchd, Linux systemd timers, GitHub Actions, or a custom agent runner.

Quick Start

Repository link:

Copy the prompt:

Use the Vault Loop prompt from PROMPT.md to design a recurring update system for this knowledge base.
First ask me for source scope, privacy boundaries, runtime environment, and the first lane I want automated.
Then create the minimum viable runbook, state file, scheduler prompt, and validation gate.

For an Agent Skills compatible install, copy the skill folder:

mkdir -p ~/.codex/skills
cp -R skill/vault-loop ~/.codex/skills/vault-loop

Then ask an agent:

Use vault-loop to set up a daily knowledge-base update lane for this vault.

Repository Layout

.
├── PROMPT.md                         # canonical reusable prompt
├── skill/vault-loop/SKILL.md          # Agent Skills compatible version
├── templates/
│   ├── automation-prompt.md           # scheduler prompt template
│   ├── runbook.md                     # recurring lane runbook template
│   ├── state.yaml                     # state file template
│   └── validation-checklist.md        # completion gate template
└── scripts/validate_project.py        # repository sanity and privacy checks

Design Principles

• Use compensation windows, not date-only shortcuts.
• Write durable knowledge to object pages, not only daily run logs.
• Keep raw sources immutable when a raw layer exists.
• Separate knowledge updates from notifications and outbound actions.
• Prefer deterministic checks for completeness, health, and state advancement.
• Fail closed: partial coverage is blocked or needs-changes, not completed.
• Keep examples generic and safe for public reuse.

Minimum Completion Standard

A lane is not complete until all of these are true:

• source scope is explicit;
• privacy boundaries are explicit;
• state cursor exists;
• runbook explains recovery;
• prompt includes startup files, collection method, writeback target, validation, health check, and state advancement rules;
• validation has passed or the run is explicitly blocked with reasons.

License

MIT. See LICENSE.