fix(verification): neutralize .NET/PowerShell-flavored examples in agnostic skills

[kyle-sexton]

Summary

The verification plugin's confirm and measure skills are sold as ecosystem-agnostic, but their illustrative examples leaned .NET/C#/PowerShell, and two "how to check" cells assumed a Unix shell (wc -l) against the cross-platform "never assume Bash" contract. A non-.NET consumer was handed a stack-specific illustration as the universal path. This patch neutralizes the examples while keeping the guidance concrete and useful.

Fix

• skills/confirm/context/fix.md — reproduction-test example now uses a generic descriptive name (returns_null_when_user_is_deleted, "name it per your framework's convention") and a generic failure ("the reported failure (e.g. a null-dereference crash)") instead of UserService_ShouldReturnNull_WhenUserIsDeleted / NullReferenceException.
• skills/measure/context/metrics.md — coupling row counts "import/dependency declarations (import/require/using, package or project references)" instead of using/ProjectReference; coverage row points to "your test runner's output" instead of dotnet test; both wc -l sites are now stated shell-neutrally (wc -l on POSIX/Git Bash, Measure-Object -Line in PowerShell), mirroring the both-forms pattern established in fix: state shell requirements with a Windows path across 8 plugins #331.
• skills/measure/context/performance.md — metric "how to measure" cells point to "your test runner", "your build tool", "your platform's profiler", and "your benchmark harness", with the shell timer given for both shells (time on POSIX/Git Bash, Measure-Command in PowerShell) instead of dotnet test/dotnet-counters/BenchmarkDotNet as the only path.
• The named marketplace-plugin evidence pointers (dotnet-*:*) at the bottom of both measure files are unchanged — those are real named plugins used as evidence sources, not universal-path illustrations, and verification: .NET/C#/PowerShell-flavored examples and Unix-shell assumptions in stack-agnostic skills (low severity) #424 deliberately did not flag them.
• verification bumped 0.2.2 → 0.2.3 (docs = patch) with a CHANGELOG entry.

Verification

Gates run in this Windows environment against the changed markdown; all clean, nothing suppressed:

• markdownlint-cli2 v0.23.0 (markdownlint v0.41.0) (repo-pinned) on the four changed .md files — Summary: 0 error(s).
• typos-cli 1.44.0 on the four changed .md files — exit 0, no output.
• editorconfig-checker v3.8.0 on the changed .md files plus plugin.json — exit 0, no findings.
• plugin.json re-parsed as valid JSON; version reads 0.2.3.

Related

Closes #424. Sibling stack-bias issues in the same work-readiness sweep, same pattern: #430, #429, #428 (all testing plugin).

🤖 Generated with Claude Code

[claude]

Claude finished @kyle-sexton's task in 1m 49s —— View job

---

PR Review — fix(verification): neutralize .NET/PowerShell-flavored examples in agnostic skills

• Gather context
• Read changed files
• Review for correctness, alignment, and quality
• Post findings

---

Verdict: Approve — clean, focused, well-reasoned patch. The changes are correct and do exactly what the PR description says they do. No blocking issues.

---

Summary

This is a documentation-only patch (5 files, 26 additions / 11 deletions) that neutralizes .NET/C#/PowerShell-flavored illustrative examples in the verification plugin's confirm and measure skills. I checked each changed file against the REVIEW.md blocking criteria — none of the security/authorization/tenancy/atomicity gates apply here. Below are observations from the content review.

---

✅ What's done well

fix.md — test name neutralization (line 52)
Replacing UserService_ShouldReturnNull_WhenUserIsDeleted (xUnit/MSTest PascalCase convention) with returns_null_when_user_is_deleted (name it per your framework's convention) is the right move. The parenthetical is a good way to acknowledge that snake_case itself is just one convention (Python/Ruby/Rust), without leaving the example unlabeled.

Dual-shell pattern is applied consistently (metrics.md L14, performance.md L14)
Both wc -l / Measure-Object -Line and time / Measure-Command pairs follow the pattern established in #331. They appear in every location where a shell command was previously implied as Unix-only.

Marketplace plugin evidence pointers left intact (metrics.md L69–73, performance.md L61–66)
The dotnet-*:* plugin references at the end of both files are real named marketplace plugin pointers, not illustrative examples. Leaving them untouched is the correct decision.

Version and CHANGELOG are consistent
Patch bump (0.2.2 → 0.2.3) is correct for a docs-only change. The CHANGELOG entry accurately describes what changed and explicitly calls out what did not change (the dotnet-* plugin pointers).

---

🟡 Nit — parenthetical inside code fence

fix.md line 52:

Reproduction test: returns_null_when_user_is_deleted (name it per your framework's convention)

Embedding (name it per your framework's convention) inside the code fence is slightly unusual — code fences typically show actual output or commands, not inline meta-commentary. A first-time reader might initially read the parenthetical as part of the test name. The surrounding prose context makes the intent clear enough, but a comment-style delimiter (e.g. # ... or a prose note before the fence) would be unambiguous. Not worth a change on its own, but worth noting.

---

🟡 Nit — table cell verbosity in metrics.md

metrics.md line 14:

Line counter (wc -l on POSIX/Git Bash, Measure-Object -Line in PowerShell), git diff --stat

The "How to check" cell is now quite long. These files are AI-readable instruction documents rather than rendered markdown for end users, so this is low impact. Still, if a future edit passes through, splitting the dual-shell note to a separate prose sentence below the table would improve scannability.

---

No concerns on REVIEW.md blocking criteria

All six blocking checks (object-level auth, tenant-scoped data access, secrets, injection, audit logging, atomicity) are inapplicable — this is a documentation-only change with no code paths, data access, or external input handling.

[claude]

Claude finished @kyle-sexton's task in 1m 26s —— View job

---

PR Review — fix(verification): neutralize .NET/PowerShell-flavored examples in agnostic skills

• Gather context
• Read changed files
• Review for correctness, alignment, and quality
• Post findings

---

Verdict: Approve — the follow-up commit cleanly resolves the one nit from the prior review, and the full patch is well-scoped and correct.

---

What changed since the last review

Commit 8c06888 addresses the nit about the parenthetical meta-note inside the code fence:

-Reproduction test: returns_null_when_user_is_deleted (name it per your framework's convention)
+Reproduction test: returns_null_when_user_is_deleted  # name it per your framework's convention

The # delimiter is the right call. A bare parenthetical inside an unlanguaged code fence could be read as part of the identifier; a # reads unambiguously as a comment in the majority of languages that use snake_case test names (Python, Ruby, Rust). The fix is minimal and targeted.

---

Full patch — residual observations

CHANGELOG date field (CHANGELOG.md L6)
The Keep a Changelog spec encourages ## [0.2.3] - YYYY-MM-DD, but all prior entries in this file omit the date too (## [0.2.2], ## [0.2.1], etc.). The new entry is consistent with the established pattern — no action needed.

git show <base>:<file> pipe in metrics.md (L24)
The prose now reads: "line count from git show <base>:<file> piped to a line counter — wc -l on POSIX/Git Bash, Measure-Object -Line in PowerShell". This is accurate — git show is cross-platform but the pipe operator has a different form in PowerShell (| vs. |... actually the same). The wording is correct; the split is only in the downstream command. Minor but fine.

REVIEW.md blocking criteria — all six inapplicable to a docs-only patch: no object-level authorization, no tenant-scoped data, no secrets, no injection surface, no audit-log requirements, no multi-file atomic write.

---

Summary

Both commits are clean. The original neutralization was well-reasoned; the follow-up precisely resolves the single nit. Ready to merge.

[kyle-sexton]

Thanks for the review — verdict noted as Approve. Classification of the two 🟡 nits:

#	Finding	Classification	Evidence	Reacted
1	fix.md L52 — (name it per your framework's convention) parenthetical inside the code fence could be misread as part of the test name	VALID — fixed	Moved the meta-note to a trailing # comment delimiter (returns_null_when_user_is_deleted # name it per your framework's convention) so it reads unambiguously as commentary, not the identifier. Fixed in 8c06888.	👍
2	metrics.md L14 — "How to check" cell verbosity from the dual-shell note	ACKNOWLEDGED (no action)	The finding is scoped to "if a future edit passes through" and is low-impact by your own read; these are AI-readable instruction docs, not rendered end-user markdown. Splitting the dual-shell note out of the table cell would reduce the table's self-containedness, so leaving it. No behavior/clarity regression.	👍

Head is now 8c06888; all checks green, merge state CLEAN. Not merging (merge gate on).