Added the standing rule "Resolve, don't hand back" to every agent instruction file fleet-wide, with a watcher that applies it on rdmsm4x when that host returns.

[2026-09-16 17:04:52 EDT · rdmpw3275m]

Why

Rich, 2026-09-16: "too much requires my attention" / "you better start making my attention count for things you can resolve, or I will stop using claude". Then: "make this sustained fleet wide, when rdmsm4x is back online".

Scope and files

• Applied on rdmpw3275m, rdmpw3265m, rdmbair13m5, rdmbair15m5, jdmbair13m5 (17:03 EDT): ~/.claude/CLAUDE.md (v4.3 line plus a bullet after the Four True Exceptions), ~/.codex/AGENTS.md, ~/.gemini/GEMINI.md, ~/.agent-coordination/AGENT_COORDINATION.md, ~/.agent-coordination/ACTIVE-DIRECTIVES.md (appended section), and the Claude memory resolve-dont-hand-back in the -Users-richh-dev project.
• rdmsm4x: offline (updating). LaunchAgent com.eastcoastscience.directive-resolve on rdmpw3275m runs propagate.zsh every 5 min, patches the hub, re-applies anywhere the hub config sync reverts it, copies the directive folder to rdmsm4x:~/dev/fleet/ops/ (never overwriting existing files), and unloads itself after 3 consecutive runs with all 6 hosts OK.
• Tooling: rdmpw3275m:~/dev/fleet/ops/directive-resolve-dont-hand-back/ (rule.md, apply_directive.py, propagate.zsh, propagate.log)
• Fleet bus broadcast, topic fleet-directive-resolve-dont-hand-back.

Verification

• apply_directive.py tested on a scratch HOME: first run APPLIED 6, second run PRESENT 6 (idempotent). Negative test with a corrupting edit: ABORT on 4 files, rc=1, file left byte-identical.
• After apply, md5 of all 5 files is identical across the 5 reachable hosts (they had also matched each other before the change).

Undo

• Every edited file has a sibling .bak-resolve- on its host.
• launchctl bootout gui/501/com.eastcoastscience.directive-resolve

Completion (2026-09-16 17:35:23 EDT, after fleet reboot)

• Watcher patched rdmsm4x and copied the tooling folder to rdmsm4x:~/dev/fleet/ops/directive-resolve-dont-hand-back at 17:15:12 EDT; all 6 hosts OK on 3 consecutive runs; it unloaded itself at 17:25:18 EDT.
• Checked after the reboot: 6/6 hosts carry the marker in 5/5 files, and the combined hash of those files (f5d4217df762) is the same on all 6, including the hub, so the hub's sync now keeps the rule in place.
• Spent LaunchAgent plist moved to rdmpw3275m:~/dev/archive/launchagents-20260916/ so it does not reload at login.