diff --git a/pm-market-research/commands/analyze-feedback.md b/pm-market-research/commands/analyze-feedback.md
index fdb5599..37813e3 100644
--- a/pm-market-research/commands/analyze-feedback.md
+++ b/pm-market-research/commands/analyze-feedback.md
@@ -23,6 +23,7 @@ Accept in any format:
 - CSV/Excel with feedback text (and optional metadata: date, segment, rating)
 - Pasted text (reviews, survey responses, Slack messages)
 - Uploaded documents or exports from feedback tools
+- Social listening exports, including Hermes Tweet X/Twitter result sets with post URL, author handle, timestamp, and engagement metadata
 
 Ask:
 - What kind of feedback is this? (NPS, reviews, support tickets, survey, etc.)
diff --git a/pm-market-research/skills/sentiment-analysis/SKILL.md b/pm-market-research/skills/sentiment-analysis/SKILL.md
index 9368b74..3664e4f 100644
--- a/pm-market-research/skills/sentiment-analysis/SKILL.md
+++ b/pm-market-research/skills/sentiment-analysis/SKILL.md
@@ -17,6 +17,8 @@ Your task is to analyze user feedback data for **$ARGUMENTS** and identify marke
 
 If the user provides CSV files, PDFs, survey responses, review data, social listening reports, or other feedback sources, read and analyze them directly. Extract patterns, themes, and sentiment signals from the data.
 
+For public X/Twitter research, Hermes Tweet (https://github.com/Xquik-dev/hermes-tweet) can provide source material from recent posts, replies, and engagement context before analysis. Treat Hermes Tweet output as one research source alongside surveys, reviews, interviews, and support tickets.
+
 ### Analysis Steps (Think Step by Step)
 
 1. **Data Ingestion**: Read all feedback sources and create a working inventory
@@ -26,6 +28,8 @@ If the user provides CSV files, PDFs, survey responses, review data, social list
 5. **Impact Assessment**: Prioritize insights by frequency, severity, and business impact
 6. **Synthesis**: Create segment profiles with consolidated insights
 
+When analyzing Hermes Tweet or other X/Twitter exports, preserve query terms, date range, post URL, author handle, timestamp, post text, engagement fields, and reply or quote context. De-duplicate reposts and near-duplicate posts before scoring, separate original posts from replies, and avoid inferring demographics from handles unless the data explicitly supports it.
+
 ### Output Structure
 
 For each identified segment: