From purchasing
Record like/dislike/nuanced feedback against a past recommendation and update the preference profile so future suggestions improve.
npx claudepluginhub danielrosehill/claude-code-plugins --plugin purchasing# /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 -...
Share bugs, ideas, or general feedback.
Close the loop on a previous recommendation. This is the single most important command for improving recommendation quality over time.
If the user named one inline (/log-feedback "The Rest is History — Constantinople"), match it against recommendations/. Otherwise:
/recommendations --unratedAsk the user (AskUserQuestion):
If the user gives a nuanced answer ("loved the host, hated the topic"), capture both dimensions.
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>
This is the load-bearing step. Open memory/preferences.md and:
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.
Tell the user what feedback was logged and what (if anything) changed in the preference profile.