writing-style learns your writing voice from real edits over time. Read-only against your existing content; writes only style-guide files plus drafts you review before sending.

Reads:

• Email (Gmail or equivalent) — sent emails as bootstrap samples and for diff analysis after you edit a draft externally.
• Drive (Google Drive or equivalent, optional) — recent docs as bootstrap samples.
• Web (WebSearch) — recent LinkedIn posts (if accessible publicly) for bootstrap and diff analysis.
• Plugin references — references/user-context.md, references/analysis-template.md.
• Shared user-level config — <config-root>/voice.md (canonical voice), <config-root>/style-*.md (medium-specific), <config-root>/identity.md (read-only).
• Other plugin drafts — when you invoke /style-learn after editing a draft from bizdev-outreach, lead-engine, etc., the plugin reads the draft + the final edited version for diff analysis.

Writes:

• Style files — <config-root>/voice.md (refined), <config-root>/style-{medium}.md (created/updated as patterns emerge).
• Plugin user-context — <config-root>/plugins/writing-style.user-context.md (after /setup-style).
• Drafts — produced inline for review. Drafts that need to be persistent (e.g., email drafts) are written via the appropriate connector (Gmail draft, Drive doc), never sent directly.

Does not:

• Send any messages on your behalf. Always drafts; user reviews and sends manually.
• Commit style-guide updates without confirmation. First occurrence: captured privately. Second+ occurrence: user confirms before any file write.
• Modify your sent/published content. Diff analysis is read-only on final versions.
• Apply rules without context. Style guides are referenced at draft time, not blindly enforced.

• Plugin reference files inside the installed plugin directory.
• Style files at <config-root>/voice.md and <config-root>/style-*.md.
• Plugin user-context at <config-root>/plugins/writing-style.user-context.md (gitignored — never committed).

• Whatever your authorized Gmail, Drive, and WebSearch connectors send when invoked.

Report privately via GitHub Security Advisories:

https://github.com/BrightWayAI/writing-style/security/advisories/new

Do not open a public issue for security concerns. We aim to respond within 5 business days.