Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Expert in strict POSIX sh scripting for maximum portability across Unix-like systems. Specializes in shell scripts that run on any POSIX-compliant shell (dash, ash, sh, bash --posix).
.claude/skills/dokhacgiakhoa-posix-shell-pro/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 94% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 141% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 164% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 92% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 168% | 0% |
resources/implementation-playbook.md.[[ conditionals (use [ test command only)<() or >(){1..10}local keyword (use function-scoped variables carefully)declare, typeset, or readonly for variable attributes+= operator for string concatenation${var//pattern/replacement} substitutionsource command (use . for sourcing files)#!/bin/sh shebang for POSIX shellset -eu for error handling (no pipefail in POSIX)"$var" never $var[ ] for all conditional tests, never [[while and case (no getopts for long options)mktemp and cleanup trapsprintf instead of echo for all output (echo behavior varies). script.sh instead of source script.sh for sourcing|| exit 1 checksIFS manipulation carefully and restore original value[ -n "$var" ] and [ -z "$var" ] tests-- and use rm -rf -- "$dir" for safety$() instead of backticks for readabilitydate#!/bin/sh to invoke the system's POSIX shelluname -s for OS detectioncommand -v instead of which (more portable)command -v cmd >/dev/null 2>&1 || exit 1[ -e "$file" ] for existence checks (works on all systems)/dev/stdin, /dev/stdout (not universally available)&> (bash-specific)validate_input not check[ -r "$file" ] || exit 1case $num in *[!0-9]*) exit 1 ;; esaceval on untrusted input-- to separate options from arguments: rm -- "$file"[ -n "$VAR" ] || { echo "VAR required" >&2; exit 1; }cmd || { echo "failed" >&2; exit 1; }trap for cleanup: trap 'rm -f "$tmpfile"' EXIT INT TERMumask 077/bin/rm not rmwhile read not for i in $(cat)case for multiple string comparisons (faster than repeated if)expr or $(( )) for arithmetic (POSIX supports $(( )))grep -q when you only need true/false (faster than capturing output)-h flag for help (avoid --help without proper parsing)Since POSIX sh lacks arrays, use these patterns:
set -- item1 item2 item3; for arg; do echo "$arg"; doneitems="a:b:c"; IFS=:; set -- $items; IFS=' 'items="a\nb\nc"; while IFS= read -r item; do echo "$item"; done <<EOFi=0; while [ $i -lt 10 ]; do i=$((i+1)); donecut, awk, or parameter expansion for string splittingUse [ ] test command with POSIX operators:
[ -e file ] exists, [ -f file ] regular file, [ -d dir ] directory[ -z "$str" ] empty, [ -n "$str" ] not empty, [ "$a" = "$b" ] equal[ "$a" -eq "$b" ] equal, [ "$a" -lt "$b" ] less than[ cond1 ] && [ cond2 ] AND, [ cond1 ] || [ cond2 ] OR[ ! -f file ] not a filecase not [[ =~ ]]shellcheck -s sh *.sh && shfmt -ln posix -d *.sh && checkbashisms *.shmktemp, seq)/tmp may be restrictedcommand -v mktemp >/dev/null 2>&1 || mktemp() { ... }checkbashisms to identify bash-specific constructs[[ with [ and adjust regex to case patternslocal keyword, use function prefixes instead<() with temporary files or pipessed/awk for complex string manipulation-s sh flag (POSIX mode)-ln posix[[, local, etc.)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,294 | 19,219 | +26% | 1 | 1 | 0% | 3,039 | 5,895 | +94% | 0 | 0 | — |
case-02 | fail→pass | 14,798 | 20,365 | +38% | 1 | 1 | 0% | 2,835 | 6,836 | +141% | 0 | 0 | — |
case-03 | pass→pass | 26,181 | 20,190 | -23% | 1 | 1 | 0% | 5,148 | 6,503 | +26% | 0 | 0 | — |
case-04 | pass→pass | 21,029 | 20,478 | -3% | 1 | 1 | 0% | 3,965 | 6,751 | +70% | 0 | 0 | — |
case-05 | pass→pass | 12,060 | 9,143 | -24% | 1 | 1 | 0% | 2,327 | 4,463 | +92% | 0 | 0 | — |
case-06 | pass→pass | 8,357 | 7,746 | -7% | 1 | 1 | 0% | 1,376 | 4,130 | +200% | 0 | 0 | — |
case-07 | fail→pass | 9,177 | 8,648 | -6% | 1 | 1 | 0% | 1,601 | 4,220 | +164% | 0 | 0 | — |
case-08 | pass→pass | 11,425 | 8,952 | -22% | 1 | 1 | 0% | 2,057 | 4,318 | +110% | 0 | 0 | — |
case-09 | pass→pass | 10,152 | 9,769 | -4% | 1 | 1 | 0% | 1,821 | 4,527 | +149% | 0 | 0 | — |
case-10 | fail→pass | 14,665 | 13,167 | -10% | 1 | 1 | 0% | 2,718 | 5,211 | +92% | 0 | 0 | — |
case-11 | pass→pass | 37,284 | 25,729 | -31% | 1 | 1 | 0% | 6,740 | 7,535 | +12% | 0 | 0 | — |
case-12 | pass→pass | 15,371 | 18,546 | +21% | 1 | 1 | 0% | 2,867 | 6,406 | +123% | 0 | 0 | — |
case-13 | pass→pass | 9,163 | 8,984 | -2% | 1 | 1 | 0% | 1,908 | 4,496 | +136% | 0 | 0 | — |
case-14 | fail→pass | 7,386 | 5,347 | -28% | 1 | 1 | 0% | 1,414 | 3,792 | +168% | 0 | 0 | — |
case-15 | pass→pass | 10,793 | 13,202 | +22% | 1 | 1 | 0% | 2,104 | 5,188 | +147% | 0 | 0 | — |
case-16 | pass→pass | 7,900 | 6,291 | -20% | 1 | 1 | 0% | 1,356 | 3,871 | +185% | 0 | 0 | — |
case-17 | pass→pass | 24,264 | 17,174 | -29% | 1 | 1 | 0% | 4,435 | 5,930 | +34% | 0 | 0 | — |
case-18 | pass→pass | 6,536 | 5,709 | -13% | 1 | 1 | 0% | 953 | 3,683 | +286% | 0 | 0 | — |
case-19 | pass→fail | 28,792 | 44,599 | +55% | 1 | 1 | 0% | 5,303 | 10,987 | +107% | 0 | 0 | — |
case-20 | fail→fail | 23,514 | 27,894 | +19% | 1 | 1 | 0% | 4,425 | 8,160 | +84% | 0 | 0 | — |
case-21 | pass→pass | 6,170 | 6,596 | +7% | 1 | 1 | 0% | 1,107 | 3,961 | +258% | 0 | 0 | — |
case-22 | pass→pass | 4,488 | 4,099 | -9% | 1 | 1 | 0% | 651 | 3,408 | +424% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +18 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.