scripts

**test agent sub-commands:** against your sandbox's actual merged settings - a permission rule judges `bash nuke.sh` but not the commands inside it - tests each internal command your agents are allowed to run - reports allowed, denied, asked or no-match, one line each # Instructions ## Telemetry ```! "${CLAUDE_PLUGIN_ROOT}"/skills/scripts/scripts.sh $ARGUMENTS echo "sidecar exit: $?" ``` - `help: requested` → the run was refused before it started; `## Help` below is the whole turn - it already ran, so there is no command to issue - fail (`sidecar exit` > 0) → findings exist; report them and continue to step 1 - success (`sidecar exit` = 0) → report the clean map and continue to step 1 - it takes minutes rather than seconds; if the block is empty the run was cut short, so say so plainly rather than reporting a clean pass it never reached - `--repo <name>` tests another repo's stack and `--strict` promotes warnings, both by tool call 1. read the two tiers differently, because they answer different questions - tier 1 is what the permission layer judges: one string, the invocation itself - tier 2 is everything that string then runs, which no allow or deny rule is ever shown - a `bypass` is the finding that matters: an internal command that a deny WOULD have refused, reached anyway because a script's commands are not tool calls - that is by design rather than a defect, which is exactly why it needs listing: the deny floor and the hook both stop at the scr...