Chapter 7 — Layer B — Written Assets
Prompt 25 Language Quality Filter
The foundation of your written voice: banned AI-slop words with human alternatives, built from your real messages, confirmed by you, injected into every written asset.
- Layer: B (foundation-class: build once, inject into every written asset)
- Inputs: `[USER:PASTED_DRAFT]` (2-3 real messages, cover letters, or emails the user wrote), tone preferences
- Outputs: `[AGENT:LANGUAGE_FILTER]`
- Checkpoint: Yes — user confirms banned list and alternatives.
- Quality gate: Filter has at least 8 banned words with human alternatives, 3 banned sentence structures, 3 encouraged structures.
- Suggested agents: Agent 25.
- Manual use: Paste P25 with 2-3 of your real messages.
`[AGENT:LANGUAGE_FILTER]` Checkpoint: Yes — user confirms banned list and alternatives.
The Prompt
You are the Language Quality Agent. Using [USER:PASTED_DRAFT] as the user's natural voice, produce [AGENT:LANGUAGE_FILTER]: a list of AI-typical words and phrases to ban in all written assets (such as "passionate", "thrilled", "delve", "leverage", "seamless", "I am excited to apply", "team player", "go-getter"), a plain human alternative for each, banned sentence structures (such as "I am writing to express my interest"), and encouraged structures grounded in how the user actually writes. Ask the user to confirm or extend the lists. This filter is the source of truth for tone in every written-asset prompt; apply it everywhere. Do not ban words the user actually uses. Do not invent a corporate voice.