skills.rest
nguyenpv1980-wq
nguyenpv1980-wqCOMMUNITY
Explore all skills by this author →
GitHub

memory-context-poisoning-reviewer

Review persistent agent memory for cross-session poisoning paths and scoping failures.

2|Updated Jul 6, 2026
Search Tags:#rollback#provenance#tenant isolation#ttl#agent security#context review#memory poisoning
One-click install
npx skills add https://github.com/nguyenpv1980-wq/Project-Aegis --skill memory-context-poisoning-reviewer
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: memory-context-poisoning-reviewer
Source: https://github.com/nguyenpv1980-wq/Project-Aegis/tree/main/.claude/skills/memory-context-poisoning-reviewer
Command: npx skills add https://github.com/nguyenpv1980-wq/Project-Aegis --skill memory-context-poisoning-reviewer

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill reviews an agent's persistent memory and stored context to find poisoning risks that survive beyond a single conversation and later steer behavior with untrusted or malicious content.

Core Features & Use Cases

  • Memory surface mapping: Identifies every long-term store, summary flow, preference cache, and recalled context that persists across sessions.
  • Write-path review: Checks whether auto-summarization, "remember this" flows, tool outputs, or peer-agent content can be persisted without validation.
  • Recall safety review: Verifies that recalled memory is treated as data with provenance and trust markers, not as instructions or policy.
  • Scope and lifecycle controls: Evaluates tenant, user, and session scoping, plus TTL, purge, rollback, and derived-state cleanup.
  • Use case: Review an assistant that stores session summaries in a memory database and confirm that attacker-planted content cannot persist, cross tenants, or reappear as trusted truth later.

Quick Start

Review the agent's persistent memory design for poisoning paths, scoping failures, recall trust issues, and purge rollback gaps.

Frequently Asked Questions about memory-context-poisoning-reviewer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
What is agent memory poisoning and how does it persist across sessions?▼

Recalled memory must be handled as data with provenance and trust markers, never as instructions or policy. Treating recalled context as executable directives allows attacker-planted content to reappear as trusted truth and steer agent behavior.

How do I review persistent agent memory for poisoning paths and scoping failures?▼

You need validation-before-write on all write paths, per-entry provenance tracking, strict tenant and session scoping to prevent cross-tenant contamination, and active TTL, purge, rollback, and derived-state cleanup controls to remove stale or poisoned data.

Why does recalled memory need provenance and trust markers in agent security?▼

Recalled memory must be handled as data with provenance and trust markers, never as instructions or policy. Treating recalled context as executable directives allows attacker-planted content to reappear as trusted truth and steer agent behavior.

Can I use context review for multi-tenant memory stores with shared state?▼

Validation-before-write ensures that auto-summarization, tool outputs, and peer-agent content cannot be persisted without validation. Without it, untrusted content enters permanent memory and survives purge or rollback operations to corrupt future agent behavior.

What are the limitations of TTL and rollback for cleaning poisoned memory?▼

TTL and rollback alone are insufficient if derived-state cleanup is missing. Poisoned content may have already been summarized or propagated to preference caches, meaning rollback of the original entry fails to remove the corrupted derived behavior.

Related Skills

Explore More Software Engineering Skills→
tiankong0101-bytetiankong0101-byte

agent-memory

Store and retrieve persistent conversational memory across agent sessions.

Community
Intermediate
ElPootElPoot

audit-quarantine

Audit quarantine and purge reversibility for duplicated comprobantes, ORS, and receptor flows.

Community
Intermediate
maruakshaymaruakshay
4

multi-tenant-model-isolation

Identify and mitigate cross-tenant leakage in shared AI deployments.

Community
Advanced
alex-voloshin-devalex-voloshin-dev
2

memory-recall

Retrieve and synthesize historical project decisions from markdown memory stores.

Community
Advanced