/rec-log-feedback

/log-feedback

Close the loop on a previous recommendation. This is the single most important command for improving recommendation quality over time.

Steps

1. Identify the recommendation

If the user named one inline (/log-feedback "The Rest is History — Constantinople"), match it against recommendations/. Otherwise:

• Run the same listing logic as /recommendations --unrated
• Ask the user to pick (AskUserQuestion or free-text reference)

2. Capture reaction

Ask the user (AskUserQuestion):

• Reaction: loved it / liked it / mixed / didn't like it / didn't engage
• Why: free-text — the reason matters more than the verdict for future tuning

If the user gives a nuanced answer ("loved the host, hated the topic"), capture both dimensions.

3. Write the feedback file

Create feedback/YYYY-MM-DD-<rec-slug>.md with:

---
recommendation: <relative path to the rec file>
reaction: loved | liked | mixed | disliked | no-engage
date: YYYY-MM-DD
---

# <rec title>

**Why:** <user's reasoning>

**Tags to lean into:** <comma-separated>
**Tags to avoid:** <comma-separated>

4. Update the preference profile

This is the load-bearing step. Open memory/preferences.md and:

• If "loved" / "liked": add specifics to Strong likes (creator, theme, style, etc.)
• If "disliked" / "no-engage": add specifics to Strong dislikes / no-go — but only if the user's reason suggests a generalisable pattern, not a one-off
• If "mixed": update Constraints or note in Open questions / experiments
• Append a dated line to Update log describing the change

Be conservative — don't over-generalise from a single data point. If a single dislike isn't a clear pattern, leave the profile alone and just record the feedback file.

5. Confirm

Tell the user what feedback was logged and what (if anything) changed in the preference profile.