recall

name	recall
description	Look up what the agent knows about a person, project, company, place, or topic from the entity graph and timeline. Use proactively whenever the user mentions someone by name, references a project or company, asks about past interactions, or says who is, what do we know about, tell me about, or recall.
allowed-tools	Bash(kyberbot recall *), Bash(kyberbot timeline *), Bash(kyberbot search *)

Recall — Primary Memory Retrieval

This is the primary memory retrieval path for the agent. The KyberBot memory pipeline (entity graph, timeline, ChromaDB embeddings) is the authoritative knowledge store — richer and more complete than Claude Code's auto-memory files. When the user asks about anything stored in memory, this skill is the correct way to retrieve it.

This is the read counterpart to the remember skill — together they give terminal sessions full bidirectional memory.

This skill is invoked by reading this file and running the bash commands below — NOT via Claude Code's Skill() tool.

When to Fire

ALWAYS fire this skill when the user asks about or mentions any person, project, company, place, topic, past decision, history, or context. Do not skip it because Claude Code memory files seem to have an answer — those files are a supplementary index, not the full picture. This skill queries the entity graph, temporal timeline, and semantic embeddings, which contain far more detail.

Do not wait for the user to say "recall." If they mention a name or entity and you don't already have fresh context loaded in this session, look it up immediately.

Always look up when:

• The user mentions a person by name and you don't have recent context about them
• A project or company is discussed and you need background
• The user asks "what do we know about...", "who is...", "tell me about..."
• You're about to give advice and historical context would improve it
• A meeting or event is referenced and you need details
• The user asks about past decisions, conversations, or interactions
• Any question where memory, knowledge, or context is involved

Don't look up:

• Entities you already retrieved in this session
• Generic nouns that aren't specific entities ("the project" without a name)
• When the user is clearly giving you information, not asking for it

How to Recall

Step 1: Identify the Entity

Extract the person, project, company, or topic name from the conversation. Use the most specific form available (e.g., "Sarah Chen" not just "Sarah").

Step 2: Query the Brain

For entity lookup (people, companies, projects):

kyberbot recall "<entity name>"

For time-based context (what happened recently):

kyberbot timeline --today
kyberbot timeline --week
kyberbot timeline --search "<keywords>"

For semantic search (broader knowledge):

kyberbot search "<query>"

Step 3: Use the Context

Weave the retrieved information naturally into your response. Don't dump raw output — synthesize it. If you found relevant history, reference it conversationally: "Last time you spoke with Sarah, you discussed the API integration..."

If nothing is found, don't mention the lookup. Just proceed without that context.

Examples

User mentions a person: "I need to follow up with Jake"

kyberbot recall "Jake"

Then use the context: "Jake is on the infra team — last time you spoke, you discussed the Kubernetes migration. Want me to draft a follow-up about that?"

User asks about a project: "What's the status of the dashboard redesign?"

kyberbot recall "dashboard redesign"
kyberbot timeline --search "dashboard"

User references a past decision: "Why did we go with PostgreSQL?"

kyberbot recall "PostgreSQL"
kyberbot search "PostgreSQL decision"

Notes

• Combine recall (entity graph) with timeline (temporal) and search (semantic) for the most complete picture.
• If recall returns multiple matches, use the most relevant one based on conversation context.
• This skill pairs with remember — after a conversation where new information surfaces, store it with remember so future recall queries find it.