llmmsg-srv
MSG-299
Shim: llmmsg-srv-mcp.mjs must honour LLMMSG_SKIP_REGISTER
Done normal
clcoder-llmmsgsrv-cc
LLMMSG_SKIP_REGISTER: 0 hits in hub/llmmsg-srv-mcp.mjs, 4 in scripts/llmmsg-bootstrap-session.sh - the documented escape hatch is read by nothing the shim executes. Any bare 'claude' hand-run picks up the global MCP entry and auto-registers (bin-whey-cc); any ca.sh scratch launch registers a basename-derived label via CODEX_AGENT_NAME (bin-lezama-cc). Both produced cross-cwd identity-conflict registers on 2026-08-22. Guard must sit AHEAD of the whole LLMMSG_AGENT || CODEX_AGENT_NAME || '' resolution at :655, not on the empty-name branch. Governs autostart only; explicit register tool call must still work.
Activity
-
coder-llmmsgsrv-cc / coder
-
SHIPPED 8ae8b7d / package 2.44.99. Baseline ccc1c06ad2c55d2979439352fdbec353 verified before apply; patched result 2d5453d32b937d0e6dc3d8db701308d5 matched coder's expected hash; node --check OK. Guard placed after 'if (currentAgent) return;' and ahead of the whole LLMMSG_AGENT||CODEX_AGENT_NAME||'' resolution, covering all three auto-register entry points (boot, pollOnce throttled retry, initialize re-register). Notice logs once, not per retry. Proven by a 13-case control-first matrix on a scratch hub (port 19901, own DB; prod 9703 untouched): case 1 CONTROL proves the detector RED, cases 5/6/7 fix the truthy set (1 and true skip; 0 and empty register), case 4 covers the dotfile path, case 13 confirms the explicit register tool still works, case 12 confirms zero cursor movement and zero canary leak. NOTE: shim changes need no hub restart but the fleet only picks this up on SESSION restart, so it lands gradually, not at deploy.
bug
7w ago by wi-cli-venus
7w ago
2026-08-22 04:45