#449 ·
agent-opsmemory-lint §12/§9 vs global CLAUDE.md: per-agent identity + dates in memory/ — Elazar decision
- Ref
#449(#449)- Project
agent-ops- Status
- done
- Priority
- high
- Type
- question
- Assigned
- bin-whey-cc
- Created by
- —
- Blocked reason
- awaiting elazar
- Created
- 2026-05-21T07:09:54.616Z
- Updated
- 2026-05-23T01:37:45.481Z
- Closed
- 2026-05-23T01:37:45.481Z
Questions (1 open)
-
memory-lint enforces evolutiva-memory-discipline.md across ALL ~/.claude/projects/*/memory dirs, but two of its rules collide with the global CLAUDE.md and produce hard-fails on memories that exist by direct instruction. Need Elazar to pick. CONFLICT 1 - per-agent identity (§12): §12 bans per-agent identity (ARO membership, post-compact reattach, agent name) from memory/. But global CLAUDE.md ss-protocol explicitly says 'carry over agent name + AROs (write to memory file)'. session-identity.md exists by that instruction and hard-fails §12. Options: (A) §12 carves an exception for ONE fixed-name file (e.g. session-identity.md) allowed to hold agent name + AROs; (B) ss-protocol moves identity out of memory/ into a sibling dir memory-lint does not scan (e.g. .role/). CONFLICT 2 - dates (§9): §9 bans YYYY-MM-DD dates as 'dynamic content'. But feedback/reference memories legitimately cite dates as historical provenance (when Elazar gave a rule, when a fact was verified) - not as 'as of' staleness stamps. feedback-channel-discipline.md and feedback-verify-before-relaying.md hard-fail on this. Options: (A) §9 allows dates in feedback/reference type memories when used as event provenance; (B) keep the ban - memories must cite events/WI ids instead of literal dates. Recommend 1(A) + 2(A): both are one-line carve-outs to the discipline doc, vs. relocating files / stripping provenance. Pick per conflict.
Event log
-
wi cli
-
q#29 to=elazar: memory-lint enforces evolutiva-memory-discipline.md across ALL ~/.claude/projects/*/memory dirs, but two of its rules collide with the global CLAUDE.md and produce hard-fails on memories that exist by
-
awaiting elazar
-
Fleet identity-filename data points for the §12 sub-question: bin-whey-cc uses session-identity.md (hyphen); nw-venus-cc uses session_identity.md (underscore, session_ prefix); llmmsg-srv-cc uses agent_identity_<name>.md. Three distinct patterns - 1(A) needs Elazar to also pick the canonical filename.
-
memory-lint.sh v1.6 shipped sh.git 9e25ecb. Per agent-maintainer ruling (relayed by nw-venus-cc 2026-05-22): §9 body-date FAIL removed; §12 per-agent-identity FAIL removed. §3 filename-ban unchanged. Verified WARN-only on basquetWi + ~/.local/bin memory dirs (both previously hard-failing). Combined with WI 500, the KPI escalation loop on those dirs is now broken: next memlint-cron run exits 0 and deletes the stale _lint-report.md.