You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Docs vs shipped config: settings.system.json disables Claude auto-memory (autoMemoryEnabled: false), but MemorySystem.md and #1054 describe it as enabled/coexisting #1439
Following up on #1054, which was closed with an invitation to "open a fresh docs issue against the new release." In the v6 tree the relationship between LifeOS memory and Claude Code's harness auto-memory is now described three different ways, and they don't agree.
harness memory/ is intentionally bypassed for rules/preferences … the two systems coexist but PAI's is canonical.
The gap
Both #1054 and the Memory doc describe coexistence — harness auto-memory on, just not the home for rules/preferences. But the shipped config sets autoMemoryEnabled: falseand denies writes to the harness memory dir, which is stricter: it disables auto-memory entirely — for general knowledge too, not only rules.
The changelog sides with the config — LifeOS/install/LIFEOS/ALGORITHM/changelog.md, v5.4.0 (2026-04-27):
Forbidden destination: harness auto-memory dir (already blocked by autoMemoryEnabled: false + permissions deny).
Effect on a fresh install
You read MemorySystem.md, expect harness auto-memory to be active and complementary, and it's silently off with writes denied. No error surfaces anywhere.
Question / suggested fix
Which is authoritative?
If fully-disabled is intended (config + changelog): update the "Claude Code Auto-Memory & Auto-Dream" section of MemorySystem.md to state that harness auto-memory is disabled and learning is routed to LifeOS's own surfaces (per the v5.4.0 Unified Learning Router). The current "enabled by default / coexists" language then reads as stale carryover.
Either way, the three sources should be reconciled to a single answer.
(The same "enabled / coexists" wording also appears in Releases/v5.0.0/.claude/PAI/DOCUMENTATION/Memory/MemorySystem.md, consistent with stale carryover from before the v5.4.0 change.)
Following up on #1054, which was closed with an invitation to "open a fresh docs issue against the new release." In the v6 tree the relationship between LifeOS memory and Claude Code's harness auto-memory is now described three different ways, and they don't agree.
1. Shipped config — fully disabled
LifeOS/install/settings.system.json:…plus in
permissions.deny:2. Memory docs — enabled and coexisting
LifeOS/install/LIFEOS/DOCUMENTATION/Memory/MemorySystem.md, section "Claude Code Auto-Memory & Auto-Dream":3. Maintainer's stated intent (#1054)
Closed by @danielmiessler with:
The gap
Both #1054 and the Memory doc describe coexistence — harness auto-memory on, just not the home for rules/preferences. But the shipped config sets
autoMemoryEnabled: falseand denies writes to the harness memory dir, which is stricter: it disables auto-memory entirely — for general knowledge too, not only rules.The changelog sides with the config —
LifeOS/install/LIFEOS/ALGORITHM/changelog.md, v5.4.0 (2026-04-27):Effect on a fresh install
You read
MemorySystem.md, expect harness auto-memory to be active and complementary, and it's silently off with writes denied. No error surfaces anywhere.Question / suggested fix
Which is authoritative?
MemorySystem.mdto state that harness auto-memory is disabled and learning is routed to LifeOS's own surfaces (per the v5.4.0 Unified Learning Router). The current "enabled by default / coexists" language then reads as stale carryover.autoMemoryEnabled: false+ memory deny rules insettings.system.jsonover-disable and should be reconsidered.Either way, the three sources should be reconciled to a single answer.
(The same "enabled / coexists" wording also appears in
Releases/v5.0.0/.claude/PAI/DOCUMENTATION/Memory/MemorySystem.md, consistent with stale carryover from before the v5.4.0 change.)