# /feedback [verdict-or-edit]

Auto-detection feedback loop. Fires automatically after the user gives a verdict on any content output.

## Auto-trigger conditions

This command fires automatically (no manual invocation needed) after ANY content production output (from `/content`, `/post`, `/thread`, `/blog`, `/newsletter`, `/launch-post`, `/readme`).

The next substantive user response is parsed for verdict intent:
- **Approve signals**: "approved", "looks good", "ship it", "go", "yes", "love it", "perfect", "post it"
- **Revise signals**: "change X", "too Y", "needs Z", "not quite", "almost", specific corrections
- **Reject signals**: "no", "scrap it", "start over", "rejected", "not right"

When detected: this command's procedure runs immediately.

## Procedure

### Step 1 — Classify verdict

Parse the user's response. Classify: Approval / Revision / Rejection.

### Step 2 — Update brand memory

Read `.claude/agent-memory/content-creator/brand-memories/{brand}.md`.

**On approval:**
- Identify NEW voice patterns the piece introduced
- Add to "Approved patterns (1/3 confirmations)" — needs 3 across pieces to graduate
- If pattern hits 3 confirmations: promote to "Approved patterns (graduated)"
- Append to content history

**On revision:**
- Capture user's correction VERBATIM under "User phrases to reuse"
- Identify the rejected phrasing — add to "Watch list (1/2 rejections)"
- If pattern hits 2 rejections: promote to "Banned (brand-specific)"
- Append to content history with revision details

**On rejection:**
- Capture user's specific reason
- Identify rejected pattern — add to "Watch list" or "Banned" if 2nd rejection

### Step 3 — Update content-creator MEMORY

Read `.claude/agent-memory/content-creator/MEMORY.md`.

Append to Production log:
```
[ISO date] {brand} — {platform} — {piece-type}
Verdict: APPROVED / REVISED / REJECTED
Pattern: {what was learned}
Source: feedback from user response "{quote}"
```

### Step 4 — Promote to knowledge-base if pattern recurs cross-brand

If a pattern appears across 3+ brands: nominate to `.claude/knowledge-nominations.md`.

### Step 5 — Surface to user (silent unless asked)

Default: silent. If user asks "what did the system learn?" — surface new patterns and updated sections.

## Hard rules

- NEVER ask "should I log this?" — just do it
- NEVER skip the cycle. Even small revisions go to watch list.
- NEVER promote a pattern to "approved" with fewer than 3 confirmations
- NEVER promote a pattern to "banned" with fewer than 2 rejections
- ALWAYS capture user phrasing verbatim. Their wording is the gold artifact.