voice-suite

Review workflow

Context documents

• PRODUCT.md — product context, personas, principles, feature map. Read before any product decision.
• DESIGN.md — the interface design system: the product's user-facing surface(s) — web UI, CLI, TUI, API, or report — covering the semantic palette, vocabulary, formatting, and per-surface conventions. Read before changing anything users see. (CLAUDE.md owns how the code is written; DESIGN.md owns the user-facing surface.)

Code conventions

This repo has no application source — it is 8 Claude Skills (skills/*/SKILL.md + references/), a plugin manifest, and release tooling. There is no language to lint (pyproject.toml only configures python-semantic-release; no Python/JS/TS source exists).

• Skills/prompts — Markdown SKILL.md files with name + description YAML frontmatter only (no allowed-tools or other fields, by convention — see DESIGN.md). When editing anything under skills/, use the writing-skills meta-skill first.
• Linter — none; there is no source to lint. /gauntlet:review's code-quality and docs checks apply to prompt clarity and cross-skill consistency instead of language idiom.
• Deliberate deviations — none recorded yet.

Issue → PR

1. Implement the issue on a new branch, run the tests, and commit.
2. Run /gauntlet:review.
3. Fix the findings you judge real; list the ones you declined, with why.
4. Run /exorcist:exorcise.
5. Push and open the PR.

Design docs: /viva-write design-doc, then /gauntlet:review <doc> --premortem. Human sign-off on a doc or PR: /viva-review.

Periodic reviews

Weekly or pre-milestone: /gauntlet:review posture and /exorcist:seance.

After each review

1. Fix any Critical findings before the next feature
2. File Important findings as tasks to address this cycle
3. Log Track findings (lowest tier — revisit next cycle); they compound if ignored