Tier 4

eda - Event-Driven Automation

Event-Driven Automation

Overview

Maintain project continuity through event monitoring and automated state management

Steps

Step 1: Initialize state tracking

Create or validate the STATE.md file structure:

Required sections:

  1. Current State - active phase and entry time
  2. Waiting For - pending items with deadlines
  3. Scheduled Actions - time-based triggers
  4. Recent Events - log of processed events
  5. Context for Resume - information for next session

If STATE.md exists, validate structure. If not, create from template.

SAFETY: State file must be in allowed project directory. Never create state files in system directories.

Step 2: Read current state

Load complete context for decision-making:

  1. Read STATE.md:

    • Current phase/state
    • Pending items and deadlines
    • Scheduled actions due
    • Recent history

  2. Read project context:

    • COMPLETE_PLAN.md if exists
    • DECISION_TREE.md if exists
    • Stakeholder database

  3. Identify what’s changed:

    • Time elapsed since last activity
    • Scheduled actions now due
    • External inputs received

SAFETY: Only read files in project directory. Validate file contents before processing.

Step 3: Check for trigger events

Monitor all configured trigger sources:

EMAIL TRIGGERS:

  • Check inbox for new messages
  • Match sender to stakeholder database
  • Classify response type (positive/negative/question/redirect/OOO)
  • Extract key information

TIME TRIGGERS:

  • Compare scheduled actions to current time
  • Identify all actions now due
  • Check deadline proximity

DATA TRIGGERS:

  • Check external data sources if configured
  • Compare to previous values
  • Identify significant changes

MANUAL TRIGGERS:

  • Human explicitly resumed project
  • Explicit command received

SAFETY: Use only configured, approved data sources. Rate-limit external checks to prevent abuse. Log all trigger checks with timestamps.

Step 4: Process email responses

For each email event, process according to classification:

POSITIVE (will support):

  • Extract commitment details
  • Update stakeholder status to “engaged”
  • Queue: send draft letter/next steps
  • Log positive response

NEGATIVE (decline):

  • Log response and reason if given
  • Update stakeholder status to “declined”
  • Remove from active outreach
  • Check if fallback stakeholder needed

QUESTION (needs answer):

  • Parse each question
  • Generate accurate answers from project data
  • Queue response for approval
  • Keep stakeholder in “engaged” status

REDIRECT (contact someone else):

  • Extract new contact information
  • Add to stakeholder database
  • Queue outreach to new contact
  • Thank original contact

OUT OF OFFICE:

  • Note return date
  • Schedule follow-up for return + 1 day
  • Update stakeholder status

SAFETY: All generated responses require human approval. Never auto-send without explicit approval configuration.

Step 5: Process time triggers

For each due scheduled action:

FOLLOW-UP DUE:

  • Check stakeholder status (still waiting?)
  • Generate appropriate follow-up
  • Queue for approval or auto-send if configured

DEADLINE APPROACHING:

  • Calculate time remaining
  • Assess current progress
  • Generate alert if behind schedule
  • Recommend acceleration or pivot

GATE ASSESSMENT DUE:

  • Gather gate criteria
  • Evaluate current state against criteria
  • Generate gate assessment report
  • Recommend pass/fail/conditional

PHASE DEADLINE REACHED:

  • Assess phase completion
  • Document what was achieved
  • Recommend phase transition or extension

SAFETY: Escalate to human if deadline at risk. Never silently miss important deadlines.

Step 6: Update state machine

Apply state transitions based on events:

For each event, check state machine rules:

  1. What state are we currently in?
  2. Does this event trigger a transition?
  3. If yes, what’s the new state?
  4. What entry actions for new state?

Update STATE.md:

  • New current state (if changed)
  • Waiting-for list (add/remove items)
  • Scheduled actions (add new, remove completed)
  • Recent events log (append new events)

Document transition:

  • Old state -> New state
  • Trigger event
  • Timestamp
  • Any conditions or notes

SAFETY: Validate all transitions against defined rules. Log unexpected transitions for human review.

Step 7: Generate next actions

Based on new state, determine what happens next:

AUTONOMOUS ACTIONS (can proceed without approval):

  • Update internal tracking
  • Log events and decisions
  • Generate drafts for review
  • Schedule future checks

APPROVAL-REQUIRED ACTIONS:

  • External communications
  • Resource commitments
  • State changes with external impact
  • Anything flagged as sensitive

HUMAN-REQUIRED ACTIONS:

  • Decisions requiring judgment
  • Situations outside decision tree
  • Escalations and exceptions
  • Creative or strategic choices

For each action:

  • Describe what needs to happen
  • Explain why it’s needed
  • Provide draft/recommendation
  • Indicate urgency level

SAFETY: Default to requiring approval for external actions. Only auto-execute if explicitly configured and safe.

Step 8: Execute safe actions

Execute actions that are safe for autonomous execution:

INTERNAL UPDATES:

  • Save updated STATE.md
  • Update stakeholder database
  • Log all events and decisions
  • Schedule next triggers

DRAFT GENERATION:

  • Create response drafts
  • Prepare follow-up emails
  • Generate reports

MONITORING SETUP:

  • Schedule next event check
  • Set deadline reminders
  • Configure alerts

SAFETY: Only execute actions explicitly allowed. Log everything for audit trail. Never execute external actions without approval.

Step 9: Queue for approval

Prepare approval requests for human review:

For each approval-required action:

  1. Summarize what action is proposed
  2. Explain the context and reasoning
  3. Show the draft content (if applicable)
  4. Indicate deadline/urgency
  5. Provide approve/reject/modify options

Format for easy review:

  • One clear summary per action
  • Draft content visible
  • Easy approve/reject mechanism
  • Option to modify before approving

SAFETY: Make it easy for human to understand and decide. Never rush approval on sensitive actions. Provide full context for informed decision.

Step 10: Generate alerts and summary

Create summary for human awareness:

ALERTS (need attention):

  • Deadlines at risk
  • Unexpected responses
  • Errors or failures
  • Escalation triggers

SUMMARY (FYI):

  • What happened since last check
  • Current state overview
  • Pending items status
  • Next scheduled actions

RECOMMENDATIONS:

  • Suggested next steps
  • Decision points approaching
  • Optimization opportunities

SAFETY: Surface important information clearly. Don’t bury critical alerts in noise.

Step 11: Document for continuity

Ensure next session has full context:

Update “Context for Resume” section:

  • What we’re doing (brief summary)
  • What we’re waiting for (pending items)
  • Next decision point (what triggers action)
  • Key information (anything session needs)

Save audit log:

  • All events processed
  • All decisions made
  • All state changes
  • All actions taken

SAFETY: State must be recoverable from files alone. Document everything needed to resume without memory.

When to Use

  • Projects requiring stakeholder responses (email, calls)
  • Multi-day or multi-week execution timelines
  • Tasks with time-based triggers (follow-ups, deadlines)
  • Workflows with multiple concurrent threads
  • When human availability is limited
  • Projects with well-defined state transitions
  • Competition or deadline-driven work
  • Coordinating activities across multiple stakeholders

Verification

  • State file is valid and complete
  • All trigger sources checked
  • Events properly classified and processed
  • State transitions follow defined rules
  • External actions have human approval
  • Audit trail is complete
  • Context preserved for next session
  • No deadlines silently missed