skills.rest
Bitterbot-AI
Bitterbot-AIOFFICIAL
Explore all skills by this author →
Website𝕏@Bitterbot_AIGitHub

recall-before-claim

Requires a memory_search call before sending messages containing unverified factual assertions.

2.5k|415|Updated Mar 28, 2026
Search Tags:#Recall#AI Agent#memory search#fact grounding#message validation#Interceptor#assertion detection
One-click install
npx skills add https://github.com/Bitterbot-AI/bitterbot-desktop --skill recall-before-claim
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: recall-before-claim
Source: https://github.com/Bitterbot-AI/bitterbot-desktop/tree/main/skills/recall-before-claim
Command: npx skills add https://github.com/Bitterbot-AI/bitterbot-desktop --skill recall-before-claim

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

AI agents often skip consulting their memory system before making factual claims, producing ungrounded assertions that users cannot easily detect. This Skill closes that gap by deterministically forcing a memory lookup before any factual claim is sent.

Core Features & Use Cases

  • Assertion Detection: Identifies factual-assertion shapes ("X is Y", "A did B") in outgoing messages and checks whether a memory tool has fired in the current turn.
  • Deterministic Enforcement: Requires a memory_search call against the assertion's subject before the message is delivered, without relying on the LLM remembering to do it.
  • Smart Skipping: Ignores opinions, questions, and assertions already grounded by a memory tool call within the last ~30 seconds; fires at most 8 times per session.
  • Use Case: When a user asks "What did we decide about the API migration?", the agent runs memory_search first and answers from actual stored memory instead of guessing from context.

Quick Start

Enable the recall-before-claim interceptor so the agent verifies factual claims against memory before sending messages.

Frequently Asked Questions about recall-before-claim

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

FAQPage Schema
How do I make an AI agent check its memory before answering?▼

Install an interceptor that detects factual assertions in outgoing messages and requires a memory_search call before sending. This skill enforces that check deterministically rather than relying on the model to remember.

How does the interceptor detect factual assertions in messages?▼

It matches assertion shapes like "X is Y" or "A did B" in the draft message. Opinion markers such as "I think" or "maybe", and questions, are skipped so only factual claims trigger the memory check.

Does the memory check fire on every message the agent sends?▼

No. It only fires when a factual assertion is ungrounded, meaning no memory tool ran in the current turn or last ~30 seconds. It is also capped at 8 firings per session to limit overhead.

Will forcing memory_search slow down agent responses?▼

Responses to factual questions may take slightly longer because the agent runs memory_search and integrates the result before sending. Casual conversation, opinions, and questions are unaffected.

What memory tools count as grounding an assertion?▼

Calls to memory_search, deep_recall, or knowledge_graph_search count as grounding. If any of these fired recently, the assertion is considered verified and the interceptor does not trigger.

Related Skills

Explore More Software Engineering Skills→
UniquecreteUniquecrete

supermemory

Stores and retrieves long-term agent memory via Supermemory's knowledge-graph API.

Community
Intermediate
DNYoussefDNYoussef
4

when-implementing-adaptive-learning-use-reasoningbank-agentdb

Automate adaptive learning for AI agents using ReasoningBank and AgentDB.

Community
Advanced
StrategicMilkStrategicMilk

Monitoring and Alerting

Detect operational anomalies in system metrics using statistical process control.

Community
Advanced
Forexgod21Forexgod21
6

authority-gradient-check

Validate instruction source authority across a four-level gradient before execution.

Community
Advanced