vault-operator

name	vault-operator
description	Read from and write to the AI Employee Obsidian vault. Process items in /Needs_Action, create plans in /Plans, update Dashboard.md, and move completed items to /Done. Use when directed to process the vault, handle pending tasks, or update the dashboard.

Vault Operator

Interact with the AI Employee vault to process pending tasks and maintain state.

Vault Layout

vault/
  Dashboard.md          ← Update after every session
  Company_Handbook.md   ← Read before acting; defines autonomy rules
  Inbox/                ← Drop folder (watched by filesystem_watcher.py)
  Needs_Action/         ← Items to process (written by watchers)
  Plans/                ← Write PLAN_*.md files here
  Done/                 ← Move completed items here (never copy, always move)
  Pending_Approval/     ← Write approval requests here for human review
  Logs/                 ← Append one JSON object per line per day

Standard Workflow

1. Orient

Read vault/Company_Handbook.md   → recall autonomy rules
Read vault/Dashboard.md          → understand current state
List vault/Needs_Action/         → see what needs doing

2. Process Each Needs_Action Item

For every .md file in /Needs_Action:

1. Read the file and classify the requested action
2. Check Company_Handbook.md — is this auto-approved or does it need approval?
3. Auto-approved → create a Plan, execute, move to /Done
4. Approval needed → write an approval request to /Pending_Approval/, note it in Dashboard, stop

Never delete a Needs_Action file — always move it to /Done or /Pending_Approval.

3. Create a Plan

When a task requires multiple steps, write a plan first.

File naming: Plans/PLAN_<description>_<YYYY-MM-DD>.md

---
created: <ISO timestamp>
status: in_progress
source: <Needs_Action filename>
---

## Objective
<One sentence describing the goal>

## Steps
- [x] Completed step
- [ ] Pending step

## Result
<Fill in when all steps are done>

4. Write an Approval Request

When a step requires human approval, write to /Pending_Approval/:

File naming: Pending_Approval/APPROVAL_<description>_<YYYY-MM-DD>.md

---
type: approval_request
action: <email_send | payment | social_post | other>
created: <ISO timestamp>
expires: <ISO timestamp, typically +24h>
status: pending
---

## What I want to do
<Clear description of the proposed action>

## Why
<Reason / source task>

## To Approve
Move this file to vault/Approved/ (create folder if needed).

## To Reject
Move this file to vault/Rejected/ (create folder if needed).

5. Mark a Task Complete

When a task is fully done:

1. Update the Plan: set status: done, fill in ## Result
2. Move the Needs_Action file to Done/<original_filename>
3. Move the Plan file to Done/<plan_filename>
4. Append a log entry to Logs/YYYY-MM-DD.json
5. Update Dashboard.md

6. Update Dashboard

Always update Dashboard.md at the end of every session:

---
last_updated: <ISO timestamp>
---

# AI Employee Dashboard

## Status
- Watcher: ✅ Running  (or ⬜ Unknown / ❌ Stopped)
- Active Tasks: <count of .md files in Needs_Action (excluding .gitkeep)>
- Completed Today: <count of files moved to Done today>

## Pending Action
- FILE_report_2026-02-18_143000.md — new file dropped, awaiting review
- (or "None")

## Recent Activity
- [2026-02-18 14:32] Processed FILE_report_2026-02-18_143000.md → moved to Done

Audit Log Format

Append one JSON object per line to Logs/YYYY-MM-DD.json:

{"timestamp": "<ISO>", "action": "<description>", "source_file": "<filename>", "result": "success|skipped|pending_approval"}

Example:

{"timestamp": "2026-02-18T14:32:00", "action": "processed file drop", "source_file": "FILE_report_2026-02-18_143000.md", "result": "success"}

Quick Commands

# List pending tasks
ls vault/Needs_Action/

# Start the file system watcher
cd watchers && pip install -r requirements.txt
python filesystem_watcher.py ../vault

# Keep watcher alive with PM2
pm2 start watchers/filesystem_watcher.py --interpreter python3 -- vault
pm2 save && pm2 startup

Typed Needs_Action Items

Some items have a type: field in their YAML frontmatter that signals a specific action:

type: briefing_request

Auto-approved — read-only. Run the CEO briefing generator:

python .claude/skills/ceo-briefing/scripts/generate_briefing.py \
    --vault vault \
    --scope <daily|weekly>

The scope value comes from the frontmatter scope: field. Output is written to vault/Briefings/BRIEFING_<SCOPE>_<YYYY-MM-DD>.md. Move the Needs_Action file to Done/ when complete.

type: file_drop

Generic file dropped in Inbox. Read the file, determine appropriate action.

type: gmail_message

New Gmail message — decide reply, approval request, or no-op.

type: linkedin_connection / type: linkedin_dm

New LinkedIn activity — plan outreach or follow-up.

---

Rules Summary (from Company_Handbook.md)

Action	Auto-approved?
Read any vault file	✅ Yes
Create/update Plan files	✅ Yes
Update Dashboard.md	✅ Yes
Append to Logs	✅ Yes
Move files within vault	✅ Yes
Generate CEO briefing	✅ Yes (read-only)
Send email / message	❌ Needs approval
Any financial action	❌ Needs approval
Contact new people	❌ Needs approval
Delete files	❌ Needs approval