Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create and modify Argcfiles using the special syntax required. Use this when editing Argcfile.sh, @argc, or any shell script that contains `argc --argc-eval`.
.claude/skills/aiskillstore-using-argc-argcfile/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 47% | 0% |
Argc is a Bash command line framework that utilizes a special comment-driven syntax to provide a command runner and argument parser.
Here is a simple Argcfile.sh:
bash# @describe Example Argcfile # Arguments, options, and flags listed here apply to the main command and all # subcommands. # # For more information about argc, see https://github.com/sigoden/argc # @option --name Name to greet # @flag -F --foo Flag param # @option --bar Option param # @option --baz* Option param (multi-occurs) # @arg val* Positional param main() { echo foo: $argc_foo echo bar: $argc_bar echo baz: ${argc_baz[@]} echo val: ${argc_val[@]} } if ! command -v argc >/dev/null; then echo "This command requires argc. Install from https://github.com/sigoden/argc" >&2 exit 100 fi eval "$(argc --argc-eval "$0" "$@")"
Run the script with some sample arguments:
bash./example.sh -F --bar=xyz --baz a --baz b v1 v2
Argc parses these arguments and creates variables prefixed with argc_:
foo: 1
bar: xyz
baz: a b
val: v1 v2You can also run ./example.sh --help to see the automatically generated help information for your CLI.
Prefer using Bash idioms when working with argc. For example, shellcheck isn't aware of the argc_foo variables, so prefer using ${argc_foo:?} for required values and values where argc is known to provide the default. For others, use ${argc_foo:-default}, as appropriate.
Note that running argc directly will attempt to locate a file named Argcfile.sh in the current and parent directories. In this case, argc will always cd into the directory with the Argcfile before executing it, and $ARGC_PWD will be set to the directory the user ran the command from.
Comment tags are standard Bash comments prefixed with @ and a specific tag. They provide instructions to Argc for configuring your script's functionalities.
| Tag | Description | | :---------- | ------------------------------------- | | @describe | Sets the description for the command. | | @cmd | Defines a subcommand. | | @alias | Sets aliases for the subcommand. | | @arg | Defines a positional argument. | | @option | Defines an option argument. | | @flag | Defines a flag argument. | | @env | Defines an environment variable. | | @meta | Adds metadata. |
Some common forms:
# @arg va
# @arg vb! required
# @arg vc* multi-values
# @arg vd+ multi-values + required
# @arg vna <PATH> value notation
# @arg vda=a default
# @arg vdb=`_default_fn` default from bash function
# @arg vca[a|b] choices
# @arg vcb[=a|b] choices + default
# @arg vcc*[a|b] multi-values + choice
# @arg vcd+[a|b] required + multi-values + choice
# @arg vfa[`_choice_fn`] choice from bash function
# @arg vfb[?`_choice_fn`] choice from bash function + no validation
# @arg vfc*[`_choice_fn`] multi-values + choice from bash function
# @arg vfd*,[`_choice_fn`] multi-values + choice from bash function + comma-separated list
# @arg vxa~ capture all remaining args
# @arg vea $$ bind-env
# @arg veb $BE <PATH> bind-named-env
# @option --oa
# @option -b --ob short
# @option -c short only
# @option --of*, multi-occurs + comma-separated list (also supports all other patterns from @arg)
# @option --ona <PATH> value notation
# @option --onb <FILE> <FILE> two-args value notations
# @option --onc <CMD> <FILE+> unlimited-args value notations
# @option --oda=a default
# @option --odb=`_default_fn` default from bash function
# @option --oca[a|b] choice (supports all patterns from @arg)
# @option --ofa[`_choice_fn`] choice from bash function (supports all patterns from @arg)
# @option --oxa~ capture all remaining args
# @option --oea $$ bind-env
# @option --oeb $BE <PATH> bind-named-env
# @flag --fa
# @flag -b --fb short
# @flag -c short only
# @flag --fd* multi-occurs
# @flag --ea $$ bind-env
# @flag --eb $BE bind-named-env
# @env EA optional
# @env EB! required
# @env EC=true default
# @env EDA[dev|prod] choices
# @env EDB[=dev|prod] choices + defaultFor _choice_fn, it should print candidates, one per line. _choice_fn and _default_fn must be bash functions, they are not arbitrary commands. "bind-env" means the variable default comes from the environment. For the environment variables, argc does not create argc_ variables (just validates existing variables).
Most common @meta options:
@meta version 1.0.0@meta require-tools git,yqUse @describe at the top level of the script and @cmd for subcommands. The first line is the short description, and subsequent comment lines that aren't comment tags are the long description.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 19,615 | 14,971 | -24% | 1 | 1 | 0% | 2,848 | 3,565 | +25% | 0 | 0 | — |
case-02 | fail→fail | 12,080 | 12,401 | +3% | 1 | 1 | 0% | 2,347 | 3,490 | +49% | 0 | 0 | — |
case-03 | fail→pass | 19,747 | 17,377 | -12% | 1 | 1 | 0% | 2,494 | 3,912 | +57% | 0 | 0 | — |
case-04 | pass→pass | 14,458 | 10,656 | -26% | 1 | 1 | 0% | 1,826 | 3,211 | +76% | 0 | 0 | — |
case-05 | pass→pass | 12,934 | 11,479 | -11% | 1 | 1 | 0% | 1,562 | 3,000 | +92% | 0 | 0 | — |
case-06 | pass→pass | 15,799 | 14,292 | -10% | 1 | 1 | 0% | 1,349 | 3,064 | +127% | 0 | 0 | — |
case-07 | pass→pass | 14,922 | 10,106 | -32% | 1 | 1 | 0% | 1,710 | 2,407 | +41% | 0 | 0 | — |
case-08 | pass→pass | 18,629 | 13,022 | -30% | 1 | 1 | 0% | 1,623 | 2,475 | +52% | 0 | 0 | — |
case-09 | pass→pass | 13,788 | 6,043 | -56% | 1 | 1 | 0% | 1,392 | 2,574 | +85% | 0 | 0 | — |
case-10 | fail→pass | 19,878 | 15,777 | -21% | 1 | 1 | 0% | 2,454 | 3,130 | +28% | 0 | 0 | — |
case-11 | fail→pass | 14,901 | 7,960 | -47% | 1 | 1 | 0% | 1,839 | 1,875 | +2% | 0 | 0 | — |
case-12 | pass→pass | 2,577 | 2,039 | -21% | 1 | 1 | 0% | 343 | 1,855 | +441% | 0 | 0 | — |
case-13 | fail→pass | 14,399 | 9,285 | -36% | 1 | 1 | 0% | 1,566 | 2,301 | +47% | 0 | 0 | — |
case-14 | pass→pass | 9,496 | 6,821 | -28% | 1 | 1 | 0% | 735 | 1,817 | +147% | 0 | 0 | — |
case-15 | fail→pass | 5,382 | 8,253 | +53% | 1 | 1 | 0% | 837 | 2,044 | +144% | 0 | 0 | — |
case-16 | fail→pass | 21,572 | 9,850 | -54% | 1 | 1 | 0% | 2,622 | 2,409 | -8% | 0 | 0 | — |
case-17 | pass→pass | 10,269 | 8,691 | -15% | 1 | 1 | 0% | 1,761 | 2,134 | +21% | 0 | 0 | — |
case-18 | pass→pass | 21,017 | 10,509 | -50% | 1 | 1 | 0% | 2,452 | 2,513 | +2% | 0 | 0 | — |
case-19 | pass→pass | 7,001 | 2,218 | -68% | 1 | 1 | 0% | 1,216 | 1,924 | +58% | 0 | 0 | — |
case-20 | fail→fail | 9,196 | 8,252 | -10% | 1 | 1 | 0% | 693 | 2,111 | +205% | 0 | 0 | — |
case-21 | fail→pass | 15,867 | 1,916 | -88% | 1 | 1 | 0% | 896 | 1,893 | +111% | 0 | 0 | — |
case-22 | pass→fail | 19,943 | 5,071 | -75% | 1 | 1 | 0% | 2,478 | 2,368 | -4% | 0 | 0 | — |
case-23 | fail→pass | 5,928 | 3,240 | -45% | 1 | 1 | 0% | 1,036 | 2,084 | +101% | 0 | 0 | — |
case-24 | fail→pass | 20,921 | 7,638 | -63% | 1 | 1 | 0% | 3,333 | 2,028 | -39% | 0 | 0 | — |
case-25 | fail→pass | 8,182 | 7,189 | -12% | 1 | 1 | 0% | 482 | 1,816 | +277% | 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. 25 cases were attempted, and 24 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +40 percentage points is the difference between those two pass rates over the 24 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.