dory-memory

What This Skill Does

The Dory-Proof Memory System is a specialized framework designed to mitigate the inherent "amnesia" of AI agents between sessions. By establishing a rigid file-based structure for persistent state, it forces the agent to externalize its context into the filesystem. This ensures that when a session terminates and restarts, the agent does not lose track of critical tasks, ongoing decisions, or user preferences. The core philosophy is that "Text > Brain"; by offloading cognitive state into dedicated markdown files within the state/ and memory/ directories, the agent regains context instantly upon boot.

Installation

To integrate this system into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/justinhartbiz/dory-memory Ensure your workspace environment allows write access to the directory so the agent can initialize the required folders and files automatically.

Use Cases

• Complex Project Management: Maintain task lists that survive server restarts or agent crashes.
• Long-term Context Retention: Store user preferences, persona quirks, and project-specific technical constraints.
• Collaborative Workspaces: Ensure multiple agent sessions or manual interventions on the filesystem are synchronized through a "single source of truth" approach.
• Safety-Critical Operations: Use the state/HOLD.md file to prevent the agent from executing potentially harmful or unfinished tasks if context was interrupted.

Example Prompts

1. "Initialize the workspace using the Dory-Proof memory system structure."
2. "I need to remember this requirement for our project: all technical documents must be in Markdown. Please save this to MEMORY.md using the scoring rubric."
3. "Check our current active task and the hold list before we proceed with the refactor."

Tips & Limitations

• Adhere to the Boot Sequence: Always read the state files in the mandated order (HOLD, ACTIVE, DECISIONS, then MEMORY) to ensure current priorities override stale historical data.
• Scoring Discipline: Use the provided 4-axis scoring table religiously. Overloading MEMORY.md with low-impact data makes it harder for the agent to find relevant information later.
• Exact Words Matter: The primary risk of context loss is paraphrasing. By forcing the agent to log the user's exact words in ACTIVE.md, you eliminate drift that usually occurs during summarization.
• Conflict Management: Respect the hierarchy. A block in HOLD.md is an absolute stop-signal for the agent, regardless of what ACTIVE.md says.