ryoitabashi/shelly/disk-cleanup-triage

When asked to free up space (on-device storage, a build directory, a project's local caches), work outward from the safest categories to the riskiest, confirming before anything irreversible:

1. Classify before deleting: reproducible build/cache artifacts

(node_modules, build output, downloaded toolchains, log files past a retention window) versus anything that's a unique source of truth (user data, uncommitted work, a config that took manual effort to set up). Never treat the second category as safe to delete just because it's old or large.

1. Measure before acting. Report what's taking the space and how much

each candidate would free, so the human can prioritize instead of the agent guessing which cleanup is "enough."

1. Prefer clearing over deleting when a safe built-in exists — a

package manager's own cache-clean command, a build tool's own clean task — over a manual rm -rf guess at what's safe, since the tool itself knows what it's safe to regenerate.

1. **Confirm nothing uncommitted lives inside a directory before removing

it.** A node_modules or build/ directory is normally safe, but a scratch/output directory can quietly accumulate a file a human meant to keep — a quick check beats an unrecoverable mistake.

1. **Never touch anything requiring elevated/broad storage permission

without being explicit about what's being deleted and why** — broad filesystem access is a capability to use narrowly, not an invitation to sweep anything that looks unused.

1. Report the before/after (space freed, what was removed, what was

left alone and why) so the cleanup is auditable after the fact, not just a silent "done."