Step 1 of 4
Define what deserves to survive
Durable memory is a curation problem before it is a storage problem.
Define the session-summary schema for durable knowledge. Sections: OUTCOME (what was accomplished, shipped, or abandoned), WORKING RECIPES (commands, configurations, and approaches that worked, exactly as used), FAILED APPROACHES (what was tried and failed, with the failure reason - so it is not repeated), USER PREFERENCES (style, tools, constraints the user expressed), PROJECT FACTS (architecture notes, key file locations, gotchas discovered). Exclusions: greetings, reasoning meander, anything reconstructible by reading the code. Write the extraction prompt that fills this schema from a session transcript, and test it on two saved transcripts - grade whether someone could resume the project from the summary alone.
Expected after this step
A summary schema plus a tested extraction prompt.
Should not happen
- ✕Summaries written only on clean exit, so crashes lose everything
- ✕Transcript-shaped summaries that are as long as what they summarize
- ✕Never loading the summaries back, making the whole layer decorative
- ✕Unbounded accumulation until the summary store itself needs compaction
Verify before continuing
Do not move on until every check is true. The complete button stays locked until then.
Do not continue if…
- !Summaries written only on clean exit, so crashes lose everything
- !Transcript-shaped summaries that are as long as what they summarize
- !Never loading the summaries back, making the whole layer decorative
- !Unbounded accumulation until the summary store itself needs compaction
If the AI messes this up
Use this when the AI fakes progress or breaks the feature. It forces a real fix.
Summaries are bloated transcripts. Apply the test 'would this line change a future session's behavior?' - delete every line that fails it, then fix the prompt to match.