Skip to content

Add intent-guardian skill: real-time task stack with forgetting detection#116

Closed
HuilingP wants to merge 1 commit intoopenclaw:mainfrom
HuilingP:add-intent-guardian
Closed

Add intent-guardian skill: real-time task stack with forgetting detection#116
HuilingP wants to merge 1 commit intoopenclaw:mainfrom
HuilingP:add-intent-guardian

Conversation

@HuilingP
Copy link

Summary

Intent Guardian is a new skill that monitors desktop activity, maintains a real-time task stack, detects when users forget what they were doing after interruptions, and sends gentle reminders to get them back on track.

Knowledge workers are interrupted every 3 minutes on average. After each interruption, there's a high chance of forgetting the original task entirely. This skill acts as an always-on focus companion that remembers your intentions when you can't.

Key Features

  • Three-tier sensing: Zero-dependency macOS osascript polling, ActivityWatch integration, or optional screenshot + vision model analysis
  • LLM-powered intent inference: Segments window activity stream into logical tasks and infers user intent
  • Real-time task stack: Tracks active, suspended, and completed tasks with interruption context
  • Forgetting detection: Distinguishes "forgotten" from "completed/postponed/abandoned" using time signals, context-switch depth, and wandering behavior
  • Personalized thresholds: Builds a focus_profile.json over time that learns which apps/transitions cause the most forgetting for each user
  • Gentle reminders: Context-rich nudges that tell you not just what you forgot, but why you left and what you found
  • Privacy-first: All data stays local, screenshots are never persisted, working hours and excluded apps are configurable

Integration with Existing Skills

Designed to complement the ecosystem:

  • screen-monitor for richer visual context
  • daily-review / personal-analytics for long-term pattern analysis
  • deepwork-tracker for auto-starting focus timers
  • rememberall for fallback time-based reminders
  • get-focus-mode for suppressing reminders during Focus mode

Files

skills/huiling-one2x/intent-guardian/
  SKILL.md                              # Main skill definition
  _meta.json                            # Metadata
  README.md                             # Documentation
  scripts/
    sense_activity.sh                   # macOS native window polling
    sense_activitywatch.sh              # ActivityWatch API integration
    sense_screen.sh                     # Optional screenshot capture
    get_task_stack.sh                   # Read current task stack
    get_focus_profile.sh                # Read focus profile
    get_recent_activity.sh              # Get recent activity entries
    log_reminder_response.sh            # Log user feedback (training signal)
    daily_focus_report.sh               # Generate daily focus summary
    cleanup.sh                          # Clean old data
  references/
    heartbeat-integration.md            # Heartbeat + Cron setup guide
    forgetting-detection.md             # Detection algorithm details

Test Plan

  • All files are pure ASCII (no encoding issues)
  • All shell scripts are executable (chmod +x)
  • SKILL.md follows OpenClaw frontmatter format with name, version, description, trigger, tools, config
  • Scripts work on macOS (primary target); sense_activitywatch.sh and daily_focus_report.sh also support Linux
  • Community testing with real OpenClaw setups welcome

Motivation

This skill fills a gap in the current ecosystem: existing skills can see the screen (screen-monitor), track time (daily-review, deepwork-tracker), and learn patterns (personal-analytics) -- but none of them connect these capabilities into real-time intent tracking and proactive forgetting detection. Intent Guardian is the "glue skill" that orchestrates these pieces into an always-on focus companion.

Made with Cursor

…tion

Intent Guardian monitors desktop activity, maintains a real-time task stack,
detects when users forget what they were doing after interruptions, and sends
gentle reminders to get them back on track.

Key features:
- Three-tier sensing: native macOS osascript, ActivityWatch, or screenshot+vision
- LLM-powered intent inference from window activity stream
- Forgetting detection with personalized thresholds
- Long-term focus pattern learning (focus_profile.json)
- Privacy-first: all data stays local, screenshots never persisted
- Integrates with existing skills (screen-monitor, daily-review, personal-analytics)

Made-with: Cursor
@openclaw-barnacle
Copy link

Thanks for the pull request! This repository is read-only and is automatically synced from https://clawhub.ai, so we can’t accept changes here. Please make updates on the website instead.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant