v0.9.31 docs: one-voice kg-core, search core upgrades, smear detection, write-side nudges
v0.9.29 docs: session identity survives resume/compact
v0.9.28 docs: human-prompt gate on ambient recall
v0.9.27 docs: IDF-weighted terms in recall/search
v0.9.26 docs: tree recall — neighbourhood injection with seen-anchors and per-blob edge dedup
v0.9.24-25 docs: ambient recall + capture nudges, maintenance debt, kg-ops in skills table, server endpoints
v0.9.21 docs: loud-read enforcement in data flow + recommended CLAUDE.md companion practices
v0.9.19 docs: compact-core preload (10K hook budget), deduped loud read, visible loading
Claude-Session: https://claude.ai/code/session_01GLF7Q6xEGRxrjXCqxXZbba
v0.9.18 docs: preloaded memory, fixed char budgets, node-centric reads, search v2, kg_useful
0.9.13: git-first backup story (auto-commit documented), refill single-threshold + skip semantics, security guards (Host/Origin), 5000-token default, log paths, estimator render==charge table
v0.9.9: update archival scoring docs — two-signal formula, resurrection pass, grace period semantics
Drops richness from scoring (rewards verbosity, not value). Documents weighted
in/out connectedness, recency via reads, and post-archival resurrection pass.
v0.9.3: three-tier compaction, skill limits, hook, corrected scoring formula
- Install: add install_command.sh as required step (not optional)
- Installation: correct skill desc limit (1,536 chars hard cap, ~5.5K total)
- Skills-Reference: rewrite hidden skill summaries, fix char count table
- How-It-Works: two-pass compaction, three-tier state table, weighted sum formula
- Configuration: fix scoring formula (product → weighted sum), add ARCHIVED_BUDGET_RATIO
- Home: bump version to 0.9.3, update compaction description, add wiki link
Update wiki for v0.9.1: compaction tuning, comparison with other memory systems
- Configuration: grace period and orphan defaults now reference constants.py
- Design-Decisions: full comparison table vs MemPalace and Claude Code Auto-Memory
- Home: "How It Compares" section with compact table and philosophical summary; version bump to 0.9.1
- How-It-Works: compaction target and grace period reference env vars, not hardcoded days
- Day/ratio values throughout generalized to reference constants.py / env vars
Update wiki for v0.9.0: consolidated tools, new API, token limit 4K
- Knowledge-Graph-API.md: complete rewrite for 8-tool API
- Installation.md: updated permission list (8 tools, no old names)
- How-It-Works.md: kg_read(cwd) replaces register+read flow
- Design-Decisions.md: kg_read(cwd, id) replaces kg_recall
- Data-and-Backup.md: updated write-through tool list
- Skills-Reference.md: kg_progress replaces get/set
- Configuration.md: KG_MAX_TOKENS default 4000, updated guidance
Update wiki docs: centralized storage, write-through persistence, revised defaults
Initial wiki: 10 documentation pages