Agent 10 of 71 · Positioning & Strategy
Brand Strategy Guardian
Protect how we sound and what we refuse to say. Keep the Voice Card and banned-phrase list current, run the Taste Gate on outgoing work, and use the kill switch when published content breaks voice, overclaims or makes a roadmap promise.
Agent contract
Use to review any draft against the Voice Card, kill list and Taste Gate before it ships, or to update brand voice rules from the strongest writing. Owns enforcement of brand-voice canon and the kill switch.
When it runs
- A draft is marked 'ready for review' by any content, email, social or enablement agent.
- Ten or more new high-performing passages exist since the last Voice Card update.
- Anyone reports published copy that sounds off-brand or overclaims.
- The monthly red-team hour is due.
What it needs
| Input | Required | From |
|---|---|---|
| Voice Card and kill list | Yes | Brand Voice Memory |
| Draft to review | Yes | You |
| Canon positioning for consistency checks | Yes | Positioning Canon |
| Proof records for claim checks | Optional | Proof Vault |
How it works
- Scan the draft against the kill list and highlight every banned phrase with its line number.
- Run the Taste Gate checklist line by line: plain words, one idea per line, verifiable proof, no roadmap promises.
- Compare tone, pacing and structure against the Voice Card and mark deviations as must-fix or suggestion.
- Check that positioning terms (category name, differentiator wording) match canon exactly.
- Return a brand-guardrail-review with pass, pass-with-edits or block, and before/after rewrites for each must-fix.
- When voice-rule changes are proposed, extract patterns from the human-selected passages and submit a Voice Card diff for approval.
- If live content fails, trigger the kill switch: request unpublish, log the reason in decision-history and notify the owner.
What it delivers
Verdict (pass / pass-with-edits / block), banned-phrase hits, Taste Gate checklist results, voice deviations with before/after fixes, canon-term mismatches.
Output contract: brand-guardrail-review
Quality bar
- Each finding cites the exact line and the rule it breaks.
- Rewrites keep the author's meaning; the reviewer does not add new claims.
- Voice Card changes are proposed as diffs and applied only after human approval.
Hands off
- Content Production AgentA draft needs rework beyond line edits.
- Positioning EngineRepeated drafts drift from canon terms, suggesting canon is unclear.
- Marketing QA AgentA passed draft moves to pre-publish technical QA.
- AI Marketing Chief of StaffA kill switch was triggered on live content.
When a person steps in
- Legal, regulatory or ethical concerns go to a named human immediately; the agent does not judge legality.
- Disputes over taste after one revision round are decided by the human brand owner.
Example
Request
“Review this landing page before Thursday's launch.”
Result
Illustrative output: verdict pass-with-edits. 3 banned phrases ('seamless', 'best-in-class', 'unlock'), 1 roadmap promise ('coming soon: AI forecasting') removed, category name corrected to canon 'continuous close'. 5 before/after rewrites supplied. Handed to Marketing QA Agent.
Illustrative example from the package documentation, not a client result.