Instruction Anchor Guard

What This Skill Does

Instruction Anchor Guard by dalomeve acts as a persistent memory layer for OpenClaw agents, specifically designed to protect critical user constraints from being lost during context window compaction or long-running tasks. In complex AI interactions, LLMs often suffer from 'plan drift' where initial rules or non-negotiable instructions are discarded as the conversation grows. This skill forces the agent to maintain an 'Anchor Ledger,' a centralized markdown file that tracks user-defined requirements. By rehydrating an 'ANCHOR_SET' at the start of every turn and performing a 'Drift Check' before any destructive action or final response, this skill ensures that your high-priority rules, formatting preferences, and forbidden actions remain strictly enforced throughout the entire lifecycle of an agent's task.

Installation

To integrate this safety mechanism into your environment, run the following command in your terminal: clawhub install openclaw/skills/skills/dalomeve/instruction-anchor-guard Once installed, the agent will automatically begin scanning for constraint-heavy language and creating the necessary audit files in your workspace.

Use Cases

• Complex Development Cycles: Ensure that specific coding standards or strict style guidelines (e.g., 'always use functional components') are never ignored after multiple file edits.
• Long-term Research: Maintain 'never' constraints, such as 'never cite sources older than 2020,' throughout a session that spans days and involves significant memory clearing.
• Agent Governance: Enforce security policies where an agent must 'always redact PII' regardless of the complexity of the data processing task.
• Workflow Automation: Prevent drift in multi-step automation sequences where intermediate steps might otherwise cause the agent to forget the ultimate goal or a specific negative constraint.

Example Prompts

1. "From now on, it is a P0 rule that you must never include introductory filler text in your responses; jump straight to the output."
2. "Important: Always format all code blocks as TypeScript and include the file path as a header, this is a highest-priority constraint."
3. "This is a long-term project: I need you to anchor the constraint that all documentation updates must strictly follow our company's internal style guide."

Tips & Limitations

• Priority Management: Avoid overusing P0 status. If everything is P0, the agent's ability to prioritize effectively is diminished. Reserve P0 for truly non-negotiable safety or structural requirements.
• Scope Awareness: Use specific scopes like task:<id> for project-specific constraints to avoid cluttering your global anchor ledger with temporary instructions.
• Conflict Resolution: The skill is designed to prevent silent failures; if you notice the agent struggling, check the .anchors.md file in your root directory to ensure the logic isn't being superseded by newer, conflicting instructions. Always review the DRIFT_CHECK blocks if an agent denies a request, as it indicates a protection mechanism is working as intended.