Imported from z5labs/ai (plugins/audit-skill/skills/audit-skill-workspace/iteration-6/eval-2/with_skill/work/.claude/skills/word-count/SKILL.md). Install upstream with npx skills add z5labs/ai --skill word-count. Copyright stays with the author.

word-count

Produce a three-number report (word count, line count, byte count) for a single text file and write the result to a sibling .count.md file.

This skill is idempotent: re-running on the same input file overwrites the report with current numbers. Re-running with a different input file produces a separate report.

Inputs

• path (required, positional CLI arg): absolute or repo-relative path to a regular file. Validation: must exist ([ -f "$path" ]) and be a text file (test with file --brief --mime-type "$path" — --brief strips the leading <path>: prefix so the raw output is the mime type; reject if it doesn't start with text/).

Output

• A file at <path>.count.md containing three lines: words: N, lines: N, bytes: N. Overwrites any existing file at that destination — re-running is the documented refresh path.

Workflow

1. Validate input. If path doesn't exist or isn't a text file, exit with a one-line error to the user and do not write anything.
2. Compute counts. Run wc -lwc "$path". wc emits fields in a fixed order — lines, words, bytes — regardless of the flag order, so the output is <lines> <words> <bytes> <path>. Map field 1 → lines, field 2 → words, field 3 → bytes. Do NOT infer the mapping from the flag order.
3. Write the report. Overwrite <path>.count.md with the three lines above. Tell the user the report path.

When to skip

If the user asks for things this skill doesn't produce — character count (vs byte count), Unicode-aware grapheme counts, frequency analysis, language detection — defer to a more capable text-analysis tool rather than misusing this one.