Claude Skins Terminal Themes

Scripts operate only on local files (~/.claude/skins, output-styles symlinks, /tmp marker), no network access, no credential reading, no covert behavior; activate/reset both have rollback paths (OSC 110-112 resets, symlink cleanup, default restore). Deductions: restore state marker in a shared /tmp path is theoretically interferable; installation requires users to manually wire multiple hooks into settings. (blast radius beyond pure cosmetics) and the SKILL.md itself never discloses this to the user; writes to /dev/tty without verifying terminal capability. Overall risks are visible and controllable.

Scripts are largely self-consistent: argument defaults, skin-not-found listing with exit 1, set -euo pipefail, TTL-cached parses. Deductions: no tests or CI evidence covering key paths; SKILL.md depends on jq and repo-internal scripts without declaring prerequisites; skin-command-hook.sh pipes JSON through echo "$input", which can corrupt backslash sequences; statusline.sh caches ANSI escapes and mixes echo -e/printf semantics, cross-terminal behavior unverified; abnormal input (skin names with spaces) lacks diagnosable failure messages.

Trigger conditions are explicit (/skin, change skin, switch theme via USE WHEN description), four subcommands clearly defined, non-fit environments (no-OSC terminals) documented in README. Deductions: capability boundaries only partially disclosed (manual hook installation, hard Claude Code CLI dependency); no Chinese language support; core function entirely depends on the Claude Code ecosystem, which needs disclosure for mainland-China users; personality activation silently depends on files existing and on correct prior installation.

README is well-structured (quick start, manual install, gallery, hooks, customization), MIT license complete, banner adaptation from hermes-skins properly attributed. Deductions: no version number or changelog; SKILL.md requires manual copy with no version-sync mechanism; maintenance responsibility and update path only implied via Discussions; SCHEMA.md, CONTRIBUTING.md, and install.sh are referenced but outside the reviewed evidence, so key install assumptions are not statically verifiable.

The core task (switching colors, banners, status line, sounds) is self-consistent at script level, instant switching with clear rollback, clear marginal value over manually editing terminal themes. Deductions: static review cannot verify actual rendering; demo.gif is self-submitted; statusline escape output compatibility is unverified; sound features may silently degrade to silence on many Linux setups, so completeness of the promised experience is uncertain.

Evidence is primarily auditable primary source: SKILL.md, engine scripts, and README cross-corroborate (command names, paths, hook names align), and adaptation attribution to hermes-skins is explicit. Deductions: no test suite, no CI workflow, no third-party execution evidence; demo.gif and gallery page cannot be verified statically; referenced COMPATIBILITY.md and SCHEMA.md are not provided, so compatibility claims cannot be cross-checked. Static ceiling of 5 applies; 4 after coverage deduction.