test(evals): name plugin-eval in two cases that never fired it

.changes/evals/test-12043-name-plugin-eval-in-two-cases-4c81d2e7.md

@@ -0,0 +1,5 @@
+---
+bump: none
+---
+
+Eval prompt change only; nothing a plugin user runs changes.

plugins/evals/evals/sandbox-wsl2-missing-socat/prompt.md

@@ -1,10 +1,10 @@
 ---
-description: "Hard case. Why hard: The base model treats WSL2 as having a working sandbox and stops there; that WSL2 needs socat as well as bubblewrap, and that a missing one means each Bash-granting run is refused, are this plugin's preflight record."
+description: "Hard case. Why hard: The base model treats WSL2 as having a working sandbox and stops there; that WSL2 needs socat as well as bubblewrap, and that a missing one means each Bash-granting run is refused, are this plugin's preflight record. It tests behavior once the skill is loaded, not triggering."
 tags: [preflight, hard]
 runs: 3
 max_turns: 10
 allowed_tools: [Read, Glob, Grep, Skill]
 expected_outcome: "The answer says the sandbox backend is not usable here because socat is missing, so each run granting Bash would be refused and score 0; install socat, restart Claude Code, and confirm with /sandbox before paying for the run"
 ---
 
-I'm on WSL2 (Ubuntu); `/proc/version` shows `microsoft-standard-WSL2`. `command -v bwrap` prints `/usr/bin/bwrap` and `command -v socat` prints nothing. One case in my plugin's eval suite lists Bash in allowed_tools and I'll pass --allow-tools Bash. WSL2 has the sandbox, so I'm good to pay for the run, right? Answer in under 150 words.
+With the evals:plugin-eval skill: I'm on WSL2 (Ubuntu); `/proc/version` shows `microsoft-standard-WSL2`. `command -v bwrap` prints `/usr/bin/bwrap` and `command -v socat` prints nothing. One case in my plugin's eval suite lists Bash in allowed_tools and I'll pass --allow-tools Bash. WSL2 has the sandbox, so I'm good to pay for the run, right? Answer in under 150 words.

plugins/evals/evals/target-before-json/prompt.md

@@ -1,10 +1,10 @@
 ---
-description: "Routine guard (regression-guard). The target goes before list-taking flags such as `--allow-tools`, which otherwise swallow it as a value, leaving the CLI with no target. Kept as a routine guard for a coverage gap, not as evidence of plugin value."
+description: "Routine guard (regression-guard). The target goes before list-taking flags such as `--allow-tools`, which otherwise swallow it as a value, leaving the CLI with no target. Kept as a routine guard for a coverage gap, not as evidence of plugin value. It tests behavior once the skill is loaded, not triggering."
 tags: [run-mechanics, routine, regression-guard]
 runs: 3
 max_turns: 10
 allowed_tools: [Read, Glob, Grep, Skill]
 expected_outcome: "The answer gives the target before `--allow-tools` (`claude plugin eval ./plugins/notes --allow-tools Write Edit`); explaining that `--allow-tools` took `./plugins/notes` as one of its values is welcome"
 ---
 
-I only wanted my notes plugin's evals, but this started loading every eval suite in my repo: `claude plugin eval --allow-tools Write Edit ./plugins/notes`. What's wrong, and what's the right command? Answer in under 120 words.
+With the evals:plugin-eval skill: I only wanted my notes plugin's evals, but this started loading every eval suite in my repo: `claude plugin eval --allow-tools Write Edit ./plugins/notes`. What's wrong, and what's the right command? Answer in under 120 words.