feat: voice-friendly skill triggers for AquaVoice (v0.14.6.0)

[garrytan]

Summary

Say "run a security check" instead of remembering /cso. Skills now have voice-friendly trigger phrases that work with AquaVoice, Whisper, and other speech-to-text tools.

Voice triggers added to 10 skills:

• /cso: "see-so", "see so", "security review", "security check", "vulnerability scan", "run security"
• /qa: "quality check", "test the app", "run QA"
• /qa-only: "bug report", "just check for bugs"
• /design-html: "build the design", "code the mockup", "make it real"
• /codex: "code x", "code ex", "get another opinion"
• /plan-eng-review: "tech review", "technical review", "plan engineering review"
• /gstack-upgrade: "upgrade the tools", "update the tools", "gee stack upgrade", "g stack upgrade"
• /autoplan: "auto plan", "automatic review"
• /benchmark: "speed test", "check performance"
• /connect-chrome: "show me the browser"

Build pipeline:

• voice-triggers: YAML field in .tmpl frontmatter (structured, clean authoring)
• gen-skill-docs preprocesses voice triggers before transformFrontmatter, folding them into the description field and stripping the field from output
• All hosts (Claude, Codex, Factory) see the updated description

Test Coverage

Tests: 289 → 296 (+7 new voice trigger tests)

• extractVoiceTriggers() parsing: valid YAML list, empty/missing field
• processVoiceTriggers(): append to description, strip field, passthrough
• Generated output: CSO SKILL.md contains "see-so", no raw voice-triggers: in frontmatter

Pre-Landing Review

Pre-Landing Review: No issues found. Template YAML data + one parser function, no security-sensitive code.

Adversarial Review

Codex adversarial review found 4 informational items (all theoretical edge cases):

1. Unquoted YAML values not handled — by design, all templates use double quotes
2. Inline descriptions not supported — all gstack descriptions use description: | block format
3. Strip regex not scoped to frontmatter — extremely unlikely collision in practice
4. Test import side-effect — pre-existing pattern in test suite

No fixes required.

Test plan

• bun run gen:skill-docs regenerates without errors
• bun test — 296 tests pass (including 7 new voice trigger tests)
• CSO SKILL.md contains "see-so" in description
• CSO SKILL.md frontmatter does NOT contain raw voice-triggers:

🤖 Generated with Claude Code

[github-actions]

E2E Evals: ❌ FAIL

31/32 tests passed | $4.03 total cost | 12 parallel runners

Suite	Result	Status	Cost
e2e-browse	4/4	✅	$0.12
e2e-deploy	4/4	✅	$0.76
e2e-design	3/3	✅	$0.54
e2e-plan	6/7	❌	$1.43
e2e-qa-workflow	3/3	✅	$0.77
e2e-review	1/1	✅	$0.09
e2e-workflow	2/2	✅	$0.16
llm-judge	8/8	✅	$0.16

---

12x ubicloud-standard-2 (Docker: pre-baked toolchain + deps) | wall clock ≈ slowest suite

Failures

• ❌ /codex-offered-ceo-review: error_api