Expand description
Live-skill shadow detector and reporting for OpenCode.
OpenCode discovers skills from more roots than any other built-in harness:
project .opencode/skills, .claude/skills, and .agents/skills (walked
up from the dispatch cwd to the git worktree), and global
$XDG_CONFIG_HOME/opencode/skills (default ~/.config/opencode/skills),
$OPENCODE_CONFIG_DIR/skills (additive, not a replacement), the legacy
~/.opencode/skills, ~/.claude/skills, and ~/.agents/skills. The
.claude/.agents roots are a cross-harness contamination vector: a skill
installed for Claude Code or Codex is visible to OpenCode sessions by
default. A logical eval skill present in any of these sources contaminates
the control arm even though eval-magic stages its test copy under a
generated slug. Detection is deliberately best-effort: missing directories
and unreadable/malformed skills are ignored while all other roots continue
to be scanned.
Functions§
- format_
shadow_ banner - OpenCode-specific build-time banner. Empty when nothing is shadowed.
- shadow_
preflight - Detect logical eval skill names that OpenCode can also load from live sources.
- shadow_
validity_ warnings - One OpenCode-specific
validity_warningsline per shadowed skill.