context-anchor_skill

Overview

This skill helps an agent recover lost context after compaction by scanning workspace memory and context files and producing a concise "where you left off" briefing. It surfaces the current task, active context files, recent decisions, and open loops so an agent can reorient quickly. Use it at session start, after compaction, or whenever you feel lost about your work.

How this skill works

The tool scans common workspace locations for current-task notes, recent daily logs, and active context files, extracting lines that indicate decisions, TODOs, blockers, and questions. It compiles timestamps, file previews, and categorized lists (current task, active files, recent decisions, open loops) into a human-readable briefing. The default scan looks back two days but can be adjusted. It runs with standard shell utilities and requires no external dependencies.

When to use it

• At the start of a session to orient yourself quickly
• Immediately after memory compaction or a restart
• When you feel unsure what you were working on
• Before handing off work to another agent
• During daily review to surface recent decisions and open items

Best practices

• Keep a single authoritative current-task file to make recovery reliable
• Tag decisions and open loops consistently (e.g., Decision:, TODO:, ?)
• Run the anchor briefing before making new changes each session
• Adjust the days scanned to balance completeness and noise
• Store active work in a predictable context/active directory for accurate previews

Example use cases

• Waking up a freshly started agent that lost compacted memory to see the current task and next steps
• Catching up on what was decided during the last two days before continuing development
• Finding and prioritizing unresolved TODOs and blockers before a sprint
• Preparing a handoff briefing to another agent or human collaborator
• Quickly previewing active project files with timestamps to pick up where you left off