Context
Follow-up from #3351. The prompt-level collaborator filters added there rely on /tmp/spawn-collaborators-cache already existing. On a schedule-triggered run (where SPAWN_ISSUE is unset), that cache may be stale or missing entirely, because:
refactor.sh / security.sh / qa.sh source collaborator-gate.sh, but only call is_collaborator when SPAWN_ISSUE is setis_collaborator is what triggers _refresh_collaborator_cache- So on schedule runs, the cache is never warmed before Claude starts — and the jq filter in each prompt returns an empty list
Failure mode
Strictly safer than the alternative (empty list vs. prompt-injection risk), so this is not a security regression. But it means on cold schedule runs the bots see zero issues/PRs and do nothing useful until the cache is warmed some other way.
Fix
In refactor.sh, security.sh, and qa.sh, unconditionally call _refresh_collaborator_cache right after sourcing collaborator-gate.sh — before Claude spawns. Something like:
if [[ -f "${GATE_SCRIPT}" ]]; then
source "${GATE_SCRIPT}"
_refresh_collaborator_cache # warm before Claude runs
fiThis makes every run, scheduled or issue-triggered, start with a fresh (or cache-valid) collaborator list.
Related
Filed from Slack by SPA
Context
Follow-up from #3351. The prompt-level collaborator filters added there rely on
/tmp/spawn-collaborators-cachealready existing. On a schedule-triggered run (whereSPAWN_ISSUEis unset), that cache may be stale or missing entirely, because:refactor.sh/security.sh/qa.shsourcecollaborator-gate.sh, but only callis_collaboratorwhenSPAWN_ISSUEis setis_collaboratoris what triggers_refresh_collaborator_cacheFailure mode
Strictly safer than the alternative (empty list vs. prompt-injection risk), so this is not a security regression. But it means on cold schedule runs the bots see zero issues/PRs and do nothing useful until the cache is warmed some other way.
Fix
In
refactor.sh,security.sh, andqa.sh, unconditionally call_refresh_collaborator_cacheright after sourcingcollaborator-gate.sh— before Claude spawns. Something like:This makes every run, scheduled or issue-triggered, start with a fresh (or cache-valid) collaborator list.
Related
Filed from Slack by SPA