[kyle-sexton]

Refs: #9886

Summary

/user-interface:design printed ${user_config.account_tools_enabled} as the current value but gave no rule for a token that arrives unexpanded. The measurement in #9886 on Claude Code 2.1.312 shows an unset key with a manifest default reaches skill content as the literal token, so a model could read the default as off. The gotcha is recorded in docs/upstream/claude-code.md (PR #10675).

Fix

The skill now says an unexpanded token means the key's declared default, true, so account-bound tools are enabled. Same approach as #10675 for /planning:interview. One changelog fragment under .changes/user-interface/.

Verification

• scripts/check-affected.sh --uncommitted --tests: 12 passed, 7 skipped (CI-only scanners), no failures.
• Manifest default confirmed: plugins/user-interface/.claude-plugin/plugin.json declares account_tools_enabled default true.
• Left out: a live 2.1.312 re-probe, the measurement is the one cited in test(planning): probe user_config substitution of an unset default and a saved boolean in a fresh session #9886.

Related

• #9886
• fix(planning): read an unexpanded user_config token as the key's default #10675

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T16:58:54.366864Z	1928579	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed 192857913cf2631677ca0c1a9e441077e183b0a8: 0 finding(s), 0 as inline comments.

No security findings.

The change adds one clause to plugins/user-interface/skills/design/SKILL.md (Step 4) and a changelog fragment. The clause says an unexpanded ${user_config.account_tools_enabled} token means the key's default, true.

• Guard default: plugins/user-interface/.claude-plugin/plugin.json:26 already declares "default": true. The manifest is unchanged, so the default does not widen. The skill text now matches it.
• What the key gates: it only decides whether the skill routes design concerns to account-bound tools (Claude Design, Figma, axe). It does not gate any trust control, token scope, merge hold or secret. Routing to a tool still requires the tool to be installed and reachable, and the skill already tells the user they may need to sign in.
• Instruction surface: nothing is deleted or softened. The diff adds a rule and no control is removed.
• PR context: it contains no embedded directives aimed at the lane.
• Manifests: no plugin.json or marketplace.json field changed, so there are no manifest <field> lines.
• Spend: nothing in this change spends prepaid credits or starts a real charge.

critical-findings: 0

[github-actions]

claude-review reviewed 192857913cf2631677ca0c1a9e441077e183b0a8: 0 finding(s), 0 as inline comments.

No findings. The change adds one sentence to plugins/user-interface/skills/design/SKILL.md Step 4, plus a patch changelog fragment (bump: patch).

• Closing issue: The PR says Refs: #9886 and has no Closes. The context file has no closing-issue section, so there are no acceptance criteria to check.
• Default claim: I checked the PR's claim against plugins/user-interface/.claude-plugin/plugin.json. account_tools_enabled is boolean with "default": true, which matches the new text.
• Manifest: The diff doesn't touch plugin.json or marketplace.json, so no manifest <field> lines apply.
• Version bump: patch fits a fix to skill instruction text.
• Security and trust: The edited file is not on the trusted-paths list. An unexpanded token now reads as true, so account-bound tools stay enabled. That matches the declared default and widens no guard.
• Verification: I did not run anything. The PR's scripts/check-affected.sh result is author-claimed and unverified. The PR also says it did no live 2.1.312 re-probe, so the unexpanded-token behavior rests on the measurement cited in test(planning): probe user_config substitution of an unset default and a saved boolean in a fresh session #9886, which I did not check.
• Spend: The change spends nothing and has no cost hold.

[github-actions]

The pr-verify verifier failed head 192857913cf2631677ca0c1a9e441077e183b0a8. What it wrote, as data:

Summary: This PR uses `Refs: #9886`, so it has to deliver a named part of #9886 and name what remains. #9886 asks for three things: record the Claude Code version, record whether the unset-default and saved-boolean cases substitute in skill content, and file a follow-up if `planning/skills/interview/SKILL.md` contradicts the result. The PR does none of these. It changes the wording in `user-interface/skills/design/SKILL.md`, which is outside the issue's scope, and its body names no work left on #9886. The wording change itself looks correct and safe.
- PR body: a `Refs:` PR must name the remaining work on the issue. 'Left out: a live 2.1.312 re-probe' is not remaining work from #9886. Either name what is left on #9886, or re-link the PR to an issue whose scope covers this change.
- .changes/user-interface/fix-9886-account-tools-unexpanded-default-3b7d5e12.md: the changelog fragment is fine, but its file name ties the fix to #9886. Rename it to match the issue the PR is re-linked to.

[github-actions]

The pr-verify verifier failed head 192857913cf2631677ca0c1a9e441077e183b0a8. What it wrote about this file, as data:

- plugins/user-interface/skills/design/SKILL.md:82: outside the scope of #9886. The issue's done-when items are a measurement comment (now posted by the host run) and, only if the interview skill is contradicted, a follow-up issue for that skill's wording. Changing `/user-interface:design` is not one of them. To fix, open or link an issue that asks for this change, such as a follow-up from the host run's caveat about default-`true` keys, and close or reference that issue instead of #9886.
- plugins/user-interface/skills/design/SKILL.md:82: no correctness or safety problem found. Treating an unexpanded `${user_config.account_tools_enabled}` as its manifest default `true` matches the #9886 measurement (an unset key with a default reaches skill content as the literal token). Per the PR body, `plugin.json` declares the default as `true`; this was not checked against the diff.