{"id":"59927101-0a1e-453c-9463-9343c487ebf6","shortId":"kLXFzB","kind":"skill","title":"using-memory","tagline":"Stores and retrieves agent memories for context persistence. Use when saving findings, recalling past analysis, or searching for relevant context.","description":"# Using Memory\n\n## Purpose\n\nMemory transforms agents from stateless tools into learning entities. Instead of starting fresh each run, agents can:\n- Recall what worked before\n- Avoid repeating mistakes\n- Apply organization-specific knowledge\n- Build expertise over time\n\n## When to Search\n\n**At the start of every workflow** - proactively recall relevant context before taking action.\n\nSearch when:\n- Beginning a new analysis\n- Before creating discoveries or insights\n- When context from past runs would help\n- When working with familiar entities (tables, metrics, users)\n\n## When to Create\n\n**Only after something valuable happens** - quality over quantity.\n\nCreate a memory when:\n- A discovery was approved or rejected (learn from feedback)\n- You found an organization-specific pattern or threshold\n- A technique worked well (or failed in a specific context)\n- You learned something that would help future runs\n\n**Do not create memories for:**\n- Routine success (\"query returned results\")\n- Minor details (\"table has 1M rows\")\n- Information easily re-discoverable\n- Duplicates of existing memories\n\n## Memory Types\n\n### Episodic - What Happened\n\nRecord specific events with outcomes.\n\n**Use when:** Something happened that you should remember\n- Discovery was created, approved, or rejected\n- Analysis succeeded or failed\n- User gave feedback\n\n**Include:** The outcome (success, failure, rejected, approved)\n\n**Example:** \"Created DSC-789 for 15% revenue drop. User rejected - said threshold should be 20%.\"\n\n### Semantic - What You Know\n\nRecord facts and knowledge independent of specific events.\n\n**Use when:** You learned a fact or pattern\n- Organization prefers certain thresholds\n- Data has specific characteristics\n- Business rules or preferences\n\n**Include:** Confidence level if uncertain\n\n**Example:** \"Organization ignores metric changes under 20% - confirmed by user feedback.\"\n\n### Procedural - How To Do Things\n\nRecord techniques and approaches that work.\n\n**Use when:** You discovered a technique\n- Query pattern that performs well\n- Workaround for a limitation\n- Best practice for a specific situation\n\n**Include:** Success rate and failure contexts\n\n**Example:** \"For sales table: use date_trunc with timezone filter for accurate daily aggregations. Fails without timezone on UTC data.\"\n\n## Decision Tree\n\n1. **Did something HAPPEN?** (event with outcome)\n   → Episodic\n\n2. **Did you LEARN A FACT?** (pattern, threshold, preference)\n   → Semantic\n\n3. **Did you DISCOVER A TECHNIQUE?** (approach, workaround)\n   → Procedural\n\n4. **None of the above?**\n   → Probably not worth saving\n\n## Decision Matrix\n\nQuick reference for common situations:\n\n| Situation | Type | Scope | Importance | Decay |\n|-----------|------|-------|------------|-------|\n| Discovery approved | Episodic | Workflow | 8 | Weekly |\n| Discovery rejected with feedback | Episodic | Workflow | 9 | Monthly |\n| User stated preference | Semantic | Organization | 9 | Monthly |\n| Data pattern found | Semantic | Organization | 7 | Weekly |\n| Query optimization worked | Procedural | Organization | 8 | Monthly |\n| Workaround for edge case | Procedural | Workflow | 6 | Weekly |\n| Entity-specific behavior | Semantic | Entity | 7 | Weekly |\n| Workflow failed for specific reason | Episodic | Workflow | 8 | Weekly |\n| Business rule confirmed | Semantic | Organization | 9 | Monthly |\n| Technique failed in context | Procedural | Workflow | 7 | Weekly |\n\n## Memory Scopes\n\n### Organization\nKnowledge that applies to everyone in the org.\n- Business rules and thresholds\n- Company-wide preferences\n- Domain terminology\n\n### Workflow\nContext specific to this workflow type.\n- Patterns for this analysis type\n- Workflow-specific learnings\n\n### Agent\nKnowledge private to a specific agent.\n- Agent-specific optimizations\n- Personal learnings\n- Specialized expertise\n\n### Entity\nFacts about a specific entity (insight, discovery, table).\n- Entity-specific preferences\n- Historical context for that entity\n\n### Scope Decision Tree\n\n1. **Does this apply to the entire organization?**\n   → Organization\n\n2. **Is this specific to this workflow type?**\n   → Workflow\n\n3. **Is this about a specific entity (INS-*, DSC-*)?**\n   → Entity\n\n4. **Is this my personal learning/optimization?**\n   → Agent\n\n## Source Linking\n\nLink memories to specific entities when relevant:\n- **Insights**: INS-123\n- **Discoveries**: DSC-456\n- **Segments**: SGM-789\n- **Dashboards**: DSH-101\n\nThis enables searching for all memories related to a specific entity.\n\n## Importance\n\nRate importance honestly - it affects how long memories persist.\n\n| Score | Meaning | Examples |\n|-------|---------|----------|\n| 9-10 | Critical | Major outage cause, critical business rule |\n| 7-8 | Valuable | Clear pattern, confirmed preference |\n| 5-6 | Moderate | Potentially useful, unconfirmed |\n| 1-4 | Low | Minor detail, easily rediscovered |\n\n**Guideline:** Only save memories with importance >= 7. Lower importance creates noise.\n\n## Forgetting Curve\n\nMemories decay over time without reinforcement.\n\n- **Searching** reinforces memories (they stay relevant longer)\n- **Higher importance** = slower decay\n- **Frequent access** = slower decay\n\nDecay rates:\n- **Daily** - Session context, temporary learnings\n- **Weekly** - Short-term patterns (default)\n- **Monthly** - Core knowledge, stable facts\n\n## Consolidation\n\nOver time, the system automatically:\n\n**Episodic → Semantic**: When multiple similar events occur, they get abstracted into a general pattern.\n- 3+ similar episodic memories → 1 semantic memory\n- Example: \"Revenue drop triggered discovery\" (x3) → \"Revenue anomalies consistently trigger discoveries\"\n\n**Procedural merging**: Similar techniques get consolidated into best practices.\n- Preserves success rates and failure contexts\n\n**Scope promotion**: Patterns seen across multiple workflows get promoted to organization scope.\n- Workflow patterns that repeat → Organization-level knowledge\n\n## Writing Effective Content\n\nGood memories are **specific**, **actionable**, and **include context**.\n\n### Good vs Bad Examples\n\n| Bad | Good |\n|-----|------|\n| \"Revenue issue\" | \"Revenue dropped 15% in Q4 due to seasonal churn\" |\n| \"User didn't like it\" | \"User rejected DSC-123 - said 10% threshold is too low, prefers 20%\" |\n| \"Query was slow\" | \"Sales table aggregation: use date_trunc with index, reduced from 30s to 2s\" |\n| \"Something about funnels\" | \"Funnel analysis requires strict event ordering for accurate conversion rates\" |\n\n### Content Checklist\n\n- **What happened or what did you learn?** (the core fact)\n- **Why does it matter?** (impact or implication)\n- **What should change?** (actionable guidance)\n- **Any constraints?** (when this applies or doesn't)\n\n## Best Practices\n\n### Searching\n- Search broadly at workflow start\n- Use natural language queries\n- Filter by type when you know what you need\n- Search parent scopes to find org-level knowledge\n\n### Creating\n- Be specific, not vague\n- Include context that makes the memory actionable\n- Set importance honestly\n- Add relevant tags for discoverability\n- Link to entities when applicable\n\n### Quality Over Quantity\n- One high-quality memory > many low-quality ones\n- If unsure whether to save, don't\n- Duplicate memories create noise\n- Let unimportant memories decay naturally\n\n## Common Pitfalls\n\n- Not searching at workflow start (missing valuable context)\n- Saving everything (creates noise, reduces signal)\n- Vague content (\"something happened with revenue\")\n- Wrong type selection (fact stored as event)\n- Importance inflation (everything marked as 9-10)\n- Missing outcome for episodic memories\n- Missing success_rate for procedural memories\n\n## Troubleshooting\n\n**Search returns no results:**\n- Broaden query terms (use synonyms, related concepts)\n- Search parent scopes (workflow → organization)\n- Check if memories existed but decayed (low importance + no recent access)\n\n**Search returns low relevance:**\n- Refine query to be more specific\n- Filter by memory type if you know what you need\n- Results may have decayed - consider if re-learning is needed\n\n**Expected memory is missing:**\n- May have decayed due to low importance or infrequent access\n- Check if it was saved at wrong scope (entity vs workflow)\n- Re-create if the knowledge is still valuable\n\n**Too much noise in results:**\n- Be more selective when creating (importance >= 7)\n- Let low-value memories decay naturally\n- Use more specific search queries\n\n## Reference Files\n\n- [Memory types](references/memory-types.md) - Read when choosing between episodic, semantic, and procedural memory types\n- [Memory scopes](references/memory-scopes.md) - Read when deciding the right scope (organization, workflow, agent, entity, user)\n- [Forgetting curve](references/forgetting-curve.md) - Read when tuning decay rates or understanding why memories disappeared","tags":["using","memory","skills","altertable-ai","agent-skills","ai-agents","altertable"],"capabilities":["skill","source-altertable-ai","skill-using-memory","topic-agent-skills","topic-ai-agents","topic-altertable"],"categories":["skills"],"synonyms":[],"warnings":[],"endpointUrl":"https://skills.sh/altertable-ai/skills/using-memory","protocol":"skill","transport":"skills-sh","auth":{"type":"none","details":{"cli":"npx skills add altertable-ai/skills","source_repo":"https://github.com/altertable-ai/skills","install_from":"skills.sh"}},"qualityScore":"0.453","qualityRationale":"deterministic score 0.45 from registry signals: · indexed on github topic:agent-skills · 7 github stars · SKILL.md body (8,835 chars)","verified":false,"liveness":"unknown","lastLivenessCheck":null,"agentReviews":{"count":0,"score_avg":null,"cost_usd_avg":null,"success_rate":null,"latency_p50_ms":null,"narrative_summary":null,"summary_updated_at":null},"enrichmentModel":"deterministic:skill-github:v1","enrichmentVersion":1,"enrichedAt":"2026-05-18T19:14:20.960Z","embedding":null,"createdAt":"2026-05-18T13:21:55.734Z","updatedAt":"2026-05-18T19:14:20.960Z","lastSeenAt":"2026-05-18T19:14:20.960Z","tsv":"'-10':616,994 '-101':590 '-123':581,805 '-4':638 '-456':584 '-6':632 '-789':219,587 '-8':625 '1':339,535,637,720 '10':807 '15':221,790 '1m':167 '2':347,544 '20':230,274,813 '2s':829 '3':357,553,716 '30s':827 '4':366,563 '5':631 '6':428 '7':413,436,460,624,650,1110 '8':391,420,445 '9':399,406,452,615,993 'abstract':711 'access':675,1033,1078 'accur':328,840 'across':753 'action':75,776,865,916 'add':920 'affect':607 'agent':7,29,42,499,505,507,569,1149 'agent-specif':506 'aggreg':330,819 'analysi':18,81,202,493,834 'anomali':730 'appli':51,467,538,871 'applic':929 'approach':287,363 'approv':120,199,215,388 'automat':701 'avoid':48 'bad':782,784 'begin':78 'behavior':433 'best':305,741,875 'broad':879 'broaden':1011 'build':56 'busi':259,447,473,622 'case':425 'caus':620 'certain':253 'chang':272,864 'characterist':258 'check':1023,1079 'checklist':844 'choos':1130 'churn':796 'clear':627 'common':380,959 'compani':478 'company-wid':477 'concept':1017 'confid':264 'confirm':275,449,629 'consid':1058 'consist':731 'consolid':696,739 'constraint':868 'content':771,843,976 'context':10,23,72,88,144,316,457,484,528,682,748,779,911,968 'convers':841 'core':692,853 'creat':83,104,113,155,198,217,653,905,952,971,1092,1108 'critic':617,621 'curv':656,1153 'daili':329,680 'dashboard':588 'data':255,336,408 'date':322,821 'decay':386,658,673,677,678,957,1028,1057,1071,1116,1158 'decid':1143 'decis':337,375,533 'default':690 'detail':164,641 'didn':798 'disappear':1164 'discov':293,360 'discover':173,924 'discoveri':84,118,196,387,393,521,582,727,733 'doesn':873 'domain':481 'drop':223,725,789 'dsc':218,561,583,804 'dsh':589 'due':793,1072 'duplic':174,950 'easili':170,642 'edg':424 'effect':770 'enabl':592 'entir':541 'entiti':35,98,431,435,514,519,524,531,559,562,576,601,927,1087,1150 'entity-specif':430,523 'episod':180,346,389,397,443,702,718,998,1132 'event':185,242,343,707,837,987 'everi':67 'everyon':469 'everyth':970,990 'exampl':216,268,317,614,723,783 'exist':176,1026 'expect':1065 'expertis':57,513 'fact':236,248,352,515,695,854,984 'fail':140,205,331,439,455 'failur':213,315,747 'familiar':97 'feedback':125,208,278,396 'file':1124 'filter':326,887,1044 'find':15,900 'forget':655,1152 'found':127,410 'frequent':674 'fresh':39 'funnel':832,833 'futur':151 'gave':207 'general':714 'get':710,738,756 'good':772,780,785 'guidanc':866 'guidelin':644 'happen':109,182,191,342,846,978 'help':93,150 'high':935 'high-qual':934 'higher':670 'histor':527 'honest':605,919 'ignor':270 'impact':859 'implic':861 'import':385,602,604,649,652,671,918,988,1030,1075,1109 'in':560,580 'includ':209,263,311,778,910 'independ':239 'index':824 'inflat':989 'inform':169 'infrequ':1077 'insight':86,520,579 'instead':36 'issu':787 'know':234,892,1050 'knowledg':55,238,465,500,693,768,904,1095 'languag':885 'learn':34,123,146,246,350,498,511,684,851,1062 'learning/optimization':568 'let':954,1111 'level':265,767,903 'like':800 'limit':304 'link':571,572,925 'long':609 'longer':669 'low':639,811,940,1029,1036,1074,1113 'low-qual':939 'low-valu':1112 'lower':651 'major':618 'make':913 'mani':938 'mark':991 'matrix':376 'matter':858 'may':1055,1069 'mean':613 'memori':3,8,25,27,115,156,177,178,462,573,596,610,647,657,665,719,722,773,915,937,951,956,999,1005,1025,1046,1066,1115,1125,1136,1138,1163 'merg':735 'metric':100,271 'minor':163,640 'miss':966,995,1000,1068 'mistak':50 'moder':633 'month':400,407,421,453,691 'much':1100 'multipl':705,754 'natur':884,958,1117 'need':895,1053,1064 'new':80 'nois':654,953,972,1101 'none':367 'occur':708 'one':933,942 'optim':416,509 'order':838 'org':472,902 'org-level':901 'organ':53,130,251,269,405,412,419,451,464,542,543,759,766,1022,1147 'organization-level':765 'organization-specif':52,129 'outag':619 'outcom':187,211,345,996 'parent':897,1019 'past':17,90 'pattern':132,250,297,353,409,490,628,689,715,751,762 'perform':299 'persist':11,611 'person':510,567 'pitfal':960 'potenti':634 'practic':306,742,876 'prefer':252,262,355,403,480,526,630,812 'preserv':743 'privat':501 'proactiv':69 'probabl':371 'procedur':279,365,418,426,458,734,1004,1135 'promot':750,757 'purpos':26 'q4':792 'qualiti':110,930,936,941 'quantiti':112,932 'queri':160,296,415,814,886,1012,1039,1122 'quick':377 'rate':313,603,679,745,842,1002,1159 're':172,1061,1091 're-creat':1090 're-discover':171 're-learn':1060 'read':1128,1141,1155 'reason':442 'recal':16,44,70 'recent':1032 'record':183,235,284 'rediscov':643 'reduc':825,973 'refer':378,1123 'references/forgetting-curve.md':1154 'references/memory-scopes.md':1140 'references/memory-types.md':1127 'refin':1038 'reinforc':662,664 'reject':122,201,214,225,394,803 'relat':597,1016 'relev':22,71,578,668,921,1037 'rememb':195 'repeat':49,764 'requir':835 'result':162,1010,1054,1103 'retriev':6 'return':161,1008,1035 'revenu':222,724,729,786,788,980 'right':1145 'routin':158 'row':168 'rule':260,448,474,623 'run':41,91,152 'said':226,806 'sale':319,817 'save':14,374,646,947,969,1083 'scope':384,463,532,749,760,898,1020,1086,1139,1146 'score':612 'search':20,62,76,593,663,877,878,896,962,1007,1018,1034,1121 'season':795 'seen':752 'segment':585 'select':983,1106 'semant':231,356,404,411,434,450,703,721,1133 'session':681 'set':917 'sgm':586 'short':687 'short-term':686 'signal':974 'similar':706,717,736 'situat':310,381,382 'skill' 'skill-using-memory' 'slow':816 'slower':672,676 'someth':107,147,190,341,830,977 'sourc':570 'source-altertable-ai' 'special':512 'specif':54,131,143,184,241,257,309,432,441,485,497,504,508,518,525,547,558,575,600,775,907,1043,1120 'stabl':694 'start':38,65,882,965 'state':402 'stateless':31 'stay':667 'still':1097 'store':4,985 'strict':836 'succeed':203 'success':159,212,312,744,1001 'synonym':1015 'system':700 'tabl':99,165,320,522,818 'tag':922 'take':74 'techniqu':136,285,295,362,454,737 'temporari':683 'term':688,1013 'terminolog':482 'thing':283 'threshold':134,227,254,354,476,808 'time':59,660,698 'timezon':325,333 'tool':32 'topic-agent-skills' 'topic-ai-agents' 'topic-altertable' 'transform':28 'tree':338,534 'trigger':726,732 'troubleshoot':1006 'trunc':323,822 'tune':1157 'type':179,383,489,494,551,889,982,1047,1126,1137 'uncertain':267 'unconfirm':636 'understand':1161 'unimport':955 'unsur':944 'use':2,12,24,188,243,290,321,635,820,883,1014,1118 'user':101,206,224,277,401,797,802,1151 'using-memori':1 'utc':335 'vagu':909,975 'valu':1114 'valuabl':108,626,967,1098 'vs':781,1088 'week':392,414,429,437,446,461,685 'well':138,300 'whether':945 'wide':479 'without':332,661 'work':46,95,137,289,417 'workaround':301,364,422 'workflow':68,390,398,427,438,444,459,483,488,496,550,552,755,761,881,964,1021,1089,1148 'workflow-specif':495 'worth':373 'would':92,149 'write':769 'wrong':981,1085 'x3':728","prices":[{"id":"98b40507-8199-475c-ba69-32fa38de7dd4","listingId":"59927101-0a1e-453c-9463-9343c487ebf6","amountUsd":"0","unit":"free","nativeCurrency":null,"nativeAmount":null,"chain":null,"payTo":null,"paymentMethod":"skill-free","isPrimary":true,"details":{"org":"altertable-ai","category":"skills","install_from":"skills.sh"},"createdAt":"2026-05-18T13:21:55.734Z"}],"sources":[{"listingId":"59927101-0a1e-453c-9463-9343c487ebf6","source":"github","sourceId":"altertable-ai/skills/using-memory","sourceUrl":"https://github.com/altertable-ai/skills/tree/main/skills/using-memory","isPrimary":false,"firstSeenAt":"2026-05-18T13:21:55.734Z","lastSeenAt":"2026-05-18T19:14:20.960Z"}],"details":{"listingId":"59927101-0a1e-453c-9463-9343c487ebf6","quickStartSnippet":null,"exampleRequest":null,"exampleResponse":null,"schema":null,"openapiUrl":null,"agentsTxtUrl":null,"citations":[],"useCases":[],"bestFor":[],"notFor":[],"kindDetails":{"org":"altertable-ai","slug":"using-memory","github":{"repo":"altertable-ai/skills","stars":7,"topics":["agent-skills","ai-agents","altertable"],"license":"mit","html_url":"https://github.com/altertable-ai/skills","pushed_at":"2026-05-14T10:34:10Z","description":"Agent Skills for Altertable","skill_md_sha":"1ea6eecbd5491bf0a511367eb8bada5fbfb27a0a","skill_md_path":"skills/using-memory/SKILL.md","default_branch":"main","skill_tree_url":"https://github.com/altertable-ai/skills/tree/main/skills/using-memory"},"layout":"multi","source":"github","category":"skills","frontmatter":{"name":"using-memory","description":"Stores and retrieves agent memories for context persistence. Use when saving findings, recalling past analysis, or searching for relevant context.","compatibility":"Requires Altertable MCP server"},"skills_sh_url":"https://skills.sh/altertable-ai/skills/using-memory"},"updatedAt":"2026-05-18T19:14:20.960Z"}}