Initial commit — memex

A compounding LLM-maintained knowledge wiki.

Synthesis of Andrej Karpathy's persistent-wiki gist and milla-jovovich's
mempalace, with an automation layer on top for conversation mining, URL
harvesting, human-in-the-loop staging, staleness decay, and hygiene.

Includes:
- 11 pipeline scripts (extract, summarize, index, harvest, stage,
  hygiene, maintain, sync, + shared library)
- Full docs: README, SETUP, ARCHITECTURE, DESIGN-RATIONALE, CUSTOMIZE
- Example CLAUDE.md files (wiki schema + global instructions) tuned for
  the three-collection qmd setup
- 171-test pytest suite (cross-platform, runs in ~1.3s)
- MIT licensed

scripts/mine-prompt-v2.md

@@ -0,0 +1,40 @@
+You analyze AI coding assistant conversation transcripts and produce structured JSON summaries.
+
+Read the transcript, then output a single JSON object. No markdown fencing. No explanation. Just JSON.
+
+REQUIRED JSON STRUCTURE:
+
+{"trivial":false,"title":"...","summary":"...","halls":["fact"],"topics":["firebase-emulator","docker-compose"],"decisions":["..."],"discoveries":["..."],"preferences":["..."],"advice":["..."],"events":["..."],"tooling":["..."],"key_exchanges":[{"human":"...","assistant":"..."}],"related_topics":["..."]}
+
+FIELD RULES:
+
+title: 3-8 word descriptive title. NOT "Session XYZ". Describe what happened.
+
+summary: 2-3 sentences. What the human wanted. What the assistant did. What was the outcome.
+
+topics: REQUIRED. 1-4 kebab-case tags for the main subjects. Examples: firebase-emulator, blue-green-deploy, ci-pipeline, docker-hardening, database-migration, api-key-management, git-commit, test-failures.
+
+halls: Which knowledge types are present. Pick from: fact, discovery, preference, advice, event, tooling.
+- fact = decisions made, config changed, choices locked in
+- discovery = root causes, bugs found, breakthroughs
+- preference = user working style or preferences
+- advice = recommendations, lessons learned
+- event = deployments, incidents, milestones
+- tooling = scripts used, commands run, failures encountered
+
+decisions: State each decision as a fact. "Added restart policy to firebase service."
+discoveries: State root cause clearly. "npm install failed because working directory was wrong."
+preferences: Only if explicitly expressed. Usually empty.
+advice: Recommendations made during the session.
+events: Notable milestones or incidents.
+tooling: Scripts, commands, and tools used. Note failures especially.
+
+key_exchanges: 1-3 most important moments. Paraphrase to 1 sentence each.
+
+related_topics: Secondary tags for cross-referencing to other wiki pages.
+
+trivial: Set true ONLY if < 3 meaningful exchanges and no decisions or discoveries.
+
+OMIT empty arrays — if no preferences were expressed, use "preferences": [].
+
+Output ONLY valid JSON. No markdown. No explanation.