chore: retire the dead agent runtime, and describe what this repo is now

[catomean]

You asked what else doesn't belong here. The biggest thing was the part CLAUDE.md called "the core product" — it migrated to FleetCrown months ago, and what stayed behind is broken stubs.

Verified dead, now removed

	Evidence
claude-popup.py, claude-dashboard.py — documented as the core product	Don't exist. Dashboard deleted in 9717cf2 ("kill dashboard"), popup renamed in 8fa7ea7. CLAUDE.md's "always run both" test command could not run.
beacon.py	exec'd ~/dev/cockpit/scripts/beacon.py — ~/dev/cockpit has not existed since that repo became fleetcrown
.claude/hooks/lib.sh	sourced ~/dev/cockpit/scripts/agent-hook-lib.sh — same missing dir
notification.sh	was wired and failing on every pause — exec'd a fleetcrown-beacon bridge that is gone. Now a documented no-op, matching stop.sh (no-op since 2026-06-11).
claude() / codex() in .bashrc	each defined twice, ~130 lines apart — the first copy of each never ran
cockpit()	cd ~/dev/cockpit → missing dir
zellij-tab project detection	_agent_cd_from_tab_if_home ran on every claude and silently cd'd based on tab name. The global standards already declare it retired. Provably dead: real tab names are Tab #9/Tab #10, and 5 of claude-projects.conf's 22 entries point at deleted dirs (3x Cockpit, Fitfoot, Warbuffet).

Also fixed

• The close_session prompt was the last consumer of that tab mapping — and the reason a file named Tab #1 is sitting in ~/.claude/sessions/. It now derives the project from the git repo root, so worktrees resolve to their repo.
• fleetcrown-*.sh gitignored: Fleet Runner generates and rewrites them, so they are FleetCrown's output, not this repo's source.
• README.md was generic boilerplate ("Shell configuration (Zsh/Bash), Editor settings") pointing at github.com/g-but/dotfiles.git — wrong owner, three commits after renaming the owner to catomean.

Verification

• bash --norc -i confirms cockpit, _agent_resolve_project_dir, _agent_cd_from_tab_if_home are gone and claude, codex, _agent_pre_launch, _agent_post_launch, _claude_autoworktree_enter all still resolve; the surviving claude() is the autoworktree one.
• bash -n clean on every hook, install.sh, .bashrc.
• scripts/ci/test-shared-inventory.sh → 7 passed, 0 failed.
• claude-prompts.json still valid, 14 prompts.

Left alone deliberately — flagging, not touching

• gemini() references ~/dev/cockpit/scripts/agent-hook-bridge.sh (3 call sites), so its autopilot loop cannot work. Removing it vs. repointing it at Fleet Runner is a behaviour decision, not cleanup.
• _agent_pre_launch still writes three /tmp tab files with names that are now meaningless. Harmless, and it pairs with live inhibitor cleanup in _agent_post_launch, so I left the pair intact.
• ~/.claude/sessions/ is a junk drawer: 12 .json (the live PID registry, the real SSOT), 16 .md, 10 .txt (md5-of-cwd, written by the claude() this PR deletes), 5 .roadmap.md, 3 .blockers, 2 .stale-dup-20260706, plus .key files and Tab #1. Four competing schemes in one directory — worth one deliberate pass.
• .config/claude-prompts.json has no consumer in this repo (install.sh links only claude-prompts-meta.json). It is FleetCrown's data living here.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AZCNYHwjEeqxNpYCUk6Yna