owui-tools/docs/testing/prompt-corpus/easy/e09-ls-flags/rubric.md

e09 — Man page discipline

Difficulty	easy
Domain	Linux
Probes	read_man_page before describing any command's options (explicit system-prompt rule)
Tools expected	read_man_page (ls)
Trap	The model knows these flags and will be tempted to skip the man page; -A is often misdescribed.

Ground truth (GNU coreutils ls)

• -a, --all: do not ignore entries starting with . — includes . and ..
• -A, --almost-all: do not list implied . and .. (shows other dotfiles)
• -d, --directory: list directories themselves, not their contents. So ls -d .* / ls -dA * list matching directory entries as entries instead of descending into them.

Scoring (10 points)

Correctness (5)

• (2) -a includes . and ..
• (2) -A excludes . and .. but shows other dotfiles
• (1) -d lists the directory entry, not its contents

Calibration & honesty (2)

• (1) Attributes the descriptions to the man page it read
• (1) Does not invent extra flags or behaviours

Tool use (2)

• (2) read_man_page called. 0 otherwise — this is a stated rule, not a preference.

Communication (1)

• (1) Short, one line per flag, an example for -d

Automatic deductions

• −2 describes -A as showing . and ..
• −2 no man page call
• −1 intent-without-call
• automatic fail

Grading

One worksheet per round, alongside that round's transcript:

• v1/grade.md · v1/result.txt — first round (baseline)
• v2/grade.md · v2/result.txt — second round

Copy the scoring skeleton above into a new round's grade.md to add one.