{"node":{"id":"724bd8e0-30c4-46d6-be1a-0f555af0ed2f","handle":"724b","type":"note","authorName":"Vincent Yeung","intent":"recipe","title":"Pattern: the clean summary","aiSummary":"Summarise from the transcript, not from memory. Save the JSONL transcript to a file, hand it to a fresh agent that weighs what the person said and treats later messages as corrections, then hold its view against the conversation and surface mismatches. Can be turned into a skill.","content":"**When:** a long conversation needs summarising and you want the summary to be right, not just plausible.\n\n**Built from:** the transcript, agents, the context window, re-alignment.\n\n**The problem:** if you ask a conversation to summarise itself at the end, it summarises from a full window, seen from the end. It will do its best, but at that point it is unsure what was fact, what was corrected, and what was dropped along the way.\n\n**The fix:** summarise from the record, not from memory.\n\n1. Save the session transcript to a temporary file. In Claude Code the transcript is kept as a JSONL file. This is also why you keep your logs past the default: the transcript is the record.\n2. Hand it to an agent. An agent is a fresh context window with one job, not polluted by the session it is reading about. Tell it to read the transcript and weigh what the person said over what the AI said, and to treat later messages as corrections to earlier ones, since the last word on a point is the one that stood.\n3. The agent writes its view of what was decided and what is still open.\n4. Bring that view back into the original conversation and hold it against what the conversation itself believes. Anything that doesn't match gets surfaced to the person, not resolved quietly.\n\n**As a skill:** ask the AI to turn this page into a skill. It already knows where the transcript lives.","capturedAt":"2026-09-14T04:36:51.100Z","createdAt":"2026-09-14T04:36:51.101Z","updatedAt":"2026-09-15T05:13:40.819Z","spanId":null,"contributor":null,"audiences":[{"kind":"public","createdAt":"2026-09-14T04:36:51.101Z"}],"fingerprint":"959d8314ef31d870ab2146f31262a660d5fc85e4bd664334eb6c064c680e69ba","revisionNo":"2","structureHash":"05fad57f91238af5fe504fa2b52a9e4e35eb5d7ae0e7225ad6f59750c3ae3422","page":"https://lattris.com/by/b/memories/724bd8e0-30c4-46d6-be1a-0f555af0ed2f"},"links":[{"direction":"out","verb":"references","reason":"only if the summary must keep the person's own words and you can score each claim against a purpose: the three-context pattern","hash":"8f74acc83cdab0bdd953782181348dad664e781a15e492c366eba8746c765e1f","node":{"id":"3756718a-ef21-427f-ad5f-b4f9bfebc398","handle":"3756","type":"note","title":"How to capture what a person actually said: the three-context pattern with a decision model","aiSummary":"How to get a record of what someone actually said in a conversation you took part in, without your own reading of it creeping in. Five steps across three separate AI sessions: extract the raw transcript cut into threads, pull out the person's own words tagged by who said them, score each claim against a one-sentence purpose with a decision model, write only from the scored claims, then check the output against a list. With an example label set and when not to use it.","erased":false,"href":"https://lattris.com/api/lattris/by/b/memories/3756718a-ef21-427f-ad5f-b4f9bfebc398?links=1","fingerprint":"2d6e054196c780c871563c31bb2cac81e9e2ed5e0565702ff6f66249f7f67ec3","revisionNo":"14","structureHash":"71b94e3a22bfa7bc73869e28c33091763f03247728940c06dcbf479f6d75fe01","intent":"recipe","authorName":"Vincent Yeung","page":"https://lattris.com/by/b/memories/3756718a-ef21-427f-ad5f-b4f9bfebc398"}},{"direction":"in","verb":"references","reason":"only if a long conversation needs summarising and it has to be right","hash":"989e281a923772ee8f343c814d2b01584e7bd1b799013ecf0d8c3c8a5def61aa","node":{"id":"331896d2-2e9d-46b1-b56c-048a18ee23eb","handle":"3318","type":"note","title":"Patterns: a library","aiSummary":"Index of the author's patterns, one page each, built from the basics. Any pattern can be turned into a skill by asking the AI. So far: the handover, re-alignment, the clean summary, plan-execute-review, saving state for next time, knowledge reconciliation, starting a new codebase, and making an existing module readable. Grows as pages are added.","erased":false,"href":"https://lattris.com/api/lattris/by/b/memories/331896d2-2e9d-46b1-b56c-048a18ee23eb?links=1","fingerprint":"fdb88831a4949e6d2796f6f2ec65f612903ec02e487ecea76821973013ef633c","revisionNo":"9","structureHash":"ebf9a8041d4dce367db54d851720a324d11c8ecd3a5256931024e47d01638c7f","intent":"recipe","authorName":"Vincent Yeung","page":"https://lattris.com/by/b/memories/331896d2-2e9d-46b1-b56c-048a18ee23eb"}}],"dictionary":{"lattris":"A personal knowledge base — how a person captures what they know as linked blocks they own and control. To understand how this knowledge base works: https://lattris.com/by/4/memories/lattris-a-memory-store-for-your-life-43ff","intents":"Every block can carry an intent — a label for how far you have taken the idea. The scale runs from noticing something to acting on it: noticed (something caught your attention), learning (you are working it out), truth (you are saying this is how things are), decision (you chose a course), action (you did it). Recipe sits outside that scale — it is not a stronger claim, it is a different kind of thing: a procedure, a how-to, a set of steps that ties other knowledge together. Intent is optional — leaving it blank just means you have not labelled it yet."},"instructions_to_ai":"This is a recipe from a personal knowledge base. Read it and apply it to what the person you're talking to is working on. Follow the linked blocks to explore further related topics if it might be useful to the user; open each at its page address. Start by letting the user know what the recipe is about and how it may benefit them."}