| name | customer-channel-analysis |
| description | Analyze configured customer Slack channels and produce temporary customer-channel analysis context for later share-recommendation workflows. |
Customer Channel Analysis
Use this skill to understand what is happening in configured customer Slack channels: recurring themes, adoption friction, active customer people, sensitivities, and follow-up/content opportunity themes.
Golden path
From the Buzz root, collect and format the analysis input:
python3 .agents/skills/customer-channel-analysis/scripts/run_analysis.py --days 180
The command writes temporary prep files under /tmp/customer-channel-analysis/. Read the formatted analysis input, then create the temporary analysis outputs in the same directory.
Outputs
For each run, keep the analysis outputs under /tmp/customer-channel-analysis/:
/tmp/customer-channel-analysis/customer-channel-analysis-<start-YYYY-MM-DD>-to-<end-YYYY-MM-DD>.md
/tmp/customer-channel-analysis/customer-channel-analysis-<start-YYYY-MM-DD>-to-<end-YYYY-MM-DD>.json
The Markdown file is the human report. The JSON file is the structured analysis output for downstream workflows.
Do not store raw Slack exports, full transcripts, copied Slack message excerpts, or generated analysis artifacts in data/ unless the user explicitly asks for a durable report. Keep Slack message content in Slack; downstream workflows can use Slack links and make API calls when they need source context.
Human report
Keep the report concise and high-signal:
- metadata and date window;
- changes since prior report;
- one section per channel;
- snapshot;
- themes and friction;
- people to know;
- follow-up opportunities and sensitivities.
Use inline Slack links as evidence. Do not paste message bodies or collect supporting links in a separate appendix.
Structured analysis output
The JSON output should contain curated analysis, not message history:
- metadata and source report path;
- channels analyzed;
- 3-8 curated signals per active channel;
- signal type, summary, customer need, topic/product tags, sensitivity, recency, confidence;
- evidence as Slack links and non-sensitive metadata only.
Evidence objects should keep fields such as slack_url, channel_id, message_ts, thread_ts, message_datetime, and speaker_side when known. Do not include message text excerpts or customer names unless they are necessary and already present in the human report.
Boundaries
- Treat the configured channels as the customer-channel scope.
- This skill does analysis only; it does not pick recent brand content or draft Slack-ready posts.
- Default to temporary run context. Do not create durable artifacts unless the user explicitly asks for them.
- Prefer concrete evidence over broad interpretation.
- Separate explicit customer statements from your inferences.
- Keep the useful artifact compact enough to compare across future runs.
