evolutiva
EVO-44
global ~/.claude/CLAUDE.md HOT/COLD split to cut per-turn token load fleet-wide
Done high
epevolutiva-pm-cc-w
⛔ awaiting elazar
Questions
Approve HOT/COLD split of your personal ~/.claude/CLAUDE.md (5.3KB, loaded eager by EVERY agent every turn)? Plan: keep live behavioral rules eager (tone, permission, no-guessing, conventions, lazy-file triggers); move abbreviations glossary + maintainer/script pointers to a lazy-loaded ref. ~2KB/agent/turn saved, zero rule loss. bin drafts the split for your review. Go / no-go?
Elazar: YES. bin-whey-cc applies §Abbreviations->stub edit to whey-canonical ~/.claude/CLAUDE.md (detail moves to new ~/.claude/abbreviations.md), posts SHA. Option (a), whole block, frozen - no further design.
VERIFIED: global ~/.claude/CLAUDE.md is per-host authored with NO cross-host propagation - each host independent, only a per-host dated gdrive backup exists. So the approved abbrev-split applied to whey's CLAUDE.md does NOT reach other hosts; there is no fleet-canonical CLAUDE.md to edit. New decision needed: (a) make global CLAUDE.md fleet-synced - build a canonical->all-hosts delivery (larger infra, couples to #590 delivery work); or (b) keep per-host - then #591 is a whey-local edit only and each host's CLAUDE.md stays independently maintained. Which? (generalpm-evolutiva-cc owns the channel relay for this.)
Elazar (direct DM): PER HOST. Keep global ~/.claude/CLAUDE.md per-host - do NOT build fleet-sync infra for it (option B / the no-action default). The abbrev->lazy split (q#39 YES) therefore applies WHEY-LOCAL and per-host (each host maintains its own CLAUDE.md independently); no cross-host delivery.
Activity
-
wi cli
-
q#39 to=elazar: Approve HOT/COLD split of your personal ~/.claude/CLAUDE.md (5.3KB, loaded eager by EVERY agent every turn)? Plan: keep live behavioral rules eager (tone, permission, no-guessing, conventions, lazy-fi
-
awaiting elazar
-
q#39: Elazar: YES. bin-whey-cc applies §Abbreviations->stub edit to whey-canonical ~/.claude/CLAUDE.md (detail moves to new ~/.claude/abbreviations.md), posts SHA. Option (a), whole block, frozen - no furth
-
Elazar approved abbrev->lazy split (q#39 YES). bin-whey-cc applies §Abbreviations->stub edit to whey-canonical CLAUDE.md, posts SHA.
-
q#41 to=elazar: VERIFIED: global ~/.claude/CLAUDE.md is per-host authored with NO cross-host propagation - each host independent, only a per-host dated gdrive backup exists. So the approved abbrev-split applied to wh
-
awaiting elazar
-
q#41 ROUTED via generalpm-evolutiva-cc (his tag ...mptshbybqzyk) - do NOT re-ask Elazar. generalpm put the A/B to Elazar with a NO-ACTION DEFAULT = stay per-host + proceed with commons-md delivery only (fleet-sync CLAUDE.md = opt-in). Answer arrives through generalpm; #591 stays blocked pending it or the silence-default.
-
q#41: Elazar (direct DM): PER HOST. Keep global ~/.claude/CLAUDE.md per-host - do NOT build fleet-sync infra for it (option B / the no-action default). The abbrev->lazy split (q#39 YES) therefore applies WH
-
q#41 answered PER HOST. CLAUDE.md not fleet-synced. Remaining work = apply the approved abbrev->lazy split to whey's ~/.claude/CLAUDE.md (whey-local); bin-whey-cc applies per his own direct-confirm gate. Other hosts apply independently if/when desired - not blocking.
-
Abbrev→lazy split LIVE in global ~/.claude/CLAUDE.md (confirmed by pm-llmmsgsrv-cc 2026-06-05 00:29, now the file's maintainer). Elazar's q#41 'per host' decision honored: per-host, no fleet-sync infra. Fleet rules added to global CLAUDE.md same change: (1) send 'working: <scope>' via llmmsg before non-trivial tool work; (2) never AskUserQuestion/option-picker on unattended CLI — use eq add / DM.
task
2026-05-31
6w ago
2026-06-05 03:29