tune-uplisted

# /tune-up — Claude Code Workflow Audit Runs a concrete, step-by-step audit of the Claude Code setup for the current project. Reports findings with a uniform per-phase output, then presents a summary with proposed actions. Does NOT apply changes without approval. ## How to execute Parse `$ARGUMENTS`. Recognised flags: - `/tune-up` — full audit; always re-fetches upstream sources - `/tune-up --offline` — skip every upstream probe (Phase 1's network leg, and the upstream half of Phases 3, 4, 5, 7). Local-only checks still run. - `/tune-up --quick` — Phase 1 (local-only) + Phase 2 + Phase 10 only. Run phases in order. For each phase, follow the numbered steps exactly, then emit the output template. After Phase 10, present the summary and wait for user approval before running Phase 11. **Reading and searching.** Prefer the dedicated tools — `Read` for files, `Grep` for content search, `Glob` for listing paths — because they go through the harness's permission and telemetry layers, while bash equivalents bypass both and obscure what the audit actually consulted. Prefer, not require: `Grep` and `Glob` are not offered to every session, and a skill that insists on a tool the session does not have cannot run at all. Check what you actually have before the first read. Where a dedicated tool is missing, use the `Bash` equivalent (`grep`, `find`, `sed -n`) and **say so once** in the Phase 1 output, so the reader knows this run's evidence came through a different path. Do not sil