An agent sandbox mounts every .git/worktrees/<id> read-only, so git worktree remove unregisters the worktree and then fails with EBUSY; container-owned files fail with EACCES. The worktree rules now forbid removal inside the sandbox and hand the operator the ownership count, chown and removal command in one message. A dirty worktree is a candidate only when all its changed and untracked content is preserved at a branch tip, and an untagged image with RepoDigests counts as a medium-risk digest pull. Both files were trimmed so the skill text does not grow. 🎯 Fix: 92 % 💥 Break: 2 % 💸 Cost: 100 % ⏱️ Time: 100 % 🛡️ Security: 100 % 🏆 Score: 90 🚦 Gate: ✅ 90/5, cost and time <= 100, security 100 % Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1.2 KiB
Docker items
Inventory
docker system df -v: containers with state and compose project, volumes
with their users (docker ps -a --filter volume=<name>), dangling and unused
images, build cache, unused networks. An untagged image with RepoDigests
is a digest-pinned pull: medium. Per stopped container its volumes
({{.Mounts}}); per volume without a running user its size, CreatedAt and
compose project, anonymous ones included.
Volumes checklist
Stopped containers and volumes get their own checklist (multiSelect: true,
header ⚠ Volumes, a further call when call 1 is full), each one unless a
peer claimed it, unknown and anonymous ones included. Question text:
Deleting a volume is irreversible; its data does not come back unless the description says how. One option per stopped container with its volumes,
one per unattached volume, anonymous ones grouped; each names size,
CreatedAt, owner or project, and what is lost.
After removing images
Built images leave their layers as build cache: re-measure docker system df, then offer docker builder prune --filter until=24h, after asking the
peers that claimed build cache, outside their builds.