From altertable
Stores and retrieves agent memories including episodic events, semantic facts, and procedural techniques for context persistence and learning. Use to save findings, recall analysis, or build knowledge across sessions.
npx claudepluginhub altertable-ai/skills --plugin altertableThis skill uses the workspace's default tool permissions.
Memory transforms agents from stateless tools into learning entities. Instead of starting fresh each run, agents can:
Guides Payload CMS config (payload.config.ts), collections, fields, hooks, access control, APIs. Debugs validation errors, security, relationships, queries, transactions, hook behavior.
Builds production-ready Apache Airflow DAGs with patterns for operators, sensors, testing, and deployment. For data pipelines, workflow orchestration, and batch jobs.
Share bugs, ideas, or general feedback.
Memory transforms agents from stateless tools into learning entities. Instead of starting fresh each run, agents can:
At the start of every workflow - proactively recall relevant context before taking action.
Search when:
Only after something valuable happens - quality over quantity.
Create a memory when:
Do not create memories for:
Record specific events with outcomes.
Use when: Something happened that you should remember
Include: The outcome (success, failure, rejected, approved)
Example: "Created DSC-789 for 15% revenue drop. User rejected - said threshold should be 20%."
Record facts and knowledge independent of specific events.
Use when: You learned a fact or pattern
Include: Confidence level if uncertain
Example: "Organization ignores metric changes under 20% - confirmed by user feedback."
Record techniques and approaches that work.
Use when: You discovered a technique
Include: Success rate and failure contexts
Example: "For sales table: use date_trunc with timezone filter for accurate daily aggregations. Fails without timezone on UTC data."
Did something HAPPEN? (event with outcome) → Episodic
Did you LEARN A FACT? (pattern, threshold, preference) → Semantic
Did you DISCOVER A TECHNIQUE? (approach, workaround) → Procedural
None of the above? → Probably not worth saving
Quick reference for common situations:
| Situation | Type | Scope | Importance | Decay |
|---|---|---|---|---|
| Discovery approved | Episodic | Workflow | 8 | Weekly |
| Discovery rejected with feedback | Episodic | Workflow | 9 | Monthly |
| User stated preference | Semantic | Organization | 9 | Monthly |
| Data pattern found | Semantic | Organization | 7 | Weekly |
| Query optimization worked | Procedural | Organization | 8 | Monthly |
| Workaround for edge case | Procedural | Workflow | 6 | Weekly |
| Entity-specific behavior | Semantic | Entity | 7 | Weekly |
| Workflow failed for specific reason | Episodic | Workflow | 8 | Weekly |
| Business rule confirmed | Semantic | Organization | 9 | Monthly |
| Technique failed in context | Procedural | Workflow | 7 | Weekly |
Knowledge that applies to everyone in the org.
Context specific to this workflow type.
Knowledge private to a specific agent.
Facts about a specific entity (insight, discovery, table).
Does this apply to the entire organization? → Organization
Is this specific to this workflow type? → Workflow
Is this about a specific entity (INS-, DSC-)? → Entity
Is this my personal learning/optimization? → Agent
Link memories to specific entities when relevant:
This enables searching for all memories related to a specific entity.
Rate importance honestly - it affects how long memories persist.
| Score | Meaning | Examples |
|---|---|---|
| 9-10 | Critical | Major outage cause, critical business rule |
| 7-8 | Valuable | Clear pattern, confirmed preference |
| 5-6 | Moderate | Potentially useful, unconfirmed |
| 1-4 | Low | Minor detail, easily rediscovered |
Guideline: Only save memories with importance >= 7. Lower importance creates noise.
Memories decay over time without reinforcement.
Decay rates:
Over time, the system automatically:
Episodic → Semantic: When multiple similar events occur, they get abstracted into a general pattern.
Procedural merging: Similar techniques get consolidated into best practices.
Scope promotion: Patterns seen across multiple workflows get promoted to organization scope.
Good memories are specific, actionable, and include context.
| Bad | Good |
|---|---|
| "Revenue issue" | "Revenue dropped 15% in Q4 due to seasonal churn" |
| "User didn't like it" | "User rejected DSC-123 - said 10% threshold is too low, prefers 20%" |
| "Query was slow" | "Sales table aggregation: use date_trunc with index, reduced from 30s to 2s" |
| "Something about funnels" | "Funnel analysis requires strict event ordering for accurate conversion rates" |
Search returns no results:
Search returns low relevance:
Expected memory is missing:
Too much noise in results: