Rats

name rats description Use when the user mentions RATS, roses and thorns, biweekly update, R&T page, biweekly retrospective, highlights and lowlights, what did I do last two weeks, confluence contribution, biweekly retro, or preparing their section for the team Roses & Thorns page. Searches Slack for the user's recent posts and produces copy-ready categorized output (Highlights, Lowlights, Learnings, Risks) matching the team R&T page format. argument-hint [--days N] allowed-tools Bash(uv run --project .claude/skills/slack python .claude/skills/slack/scripts/*.py *) Roses & Thorns -- Biweekly Contribution Automator Search Slack for your recent posts, auto-categorize them into Highlights / Lowlights / Learnings / Risks, rewrite them in your voice, and present copy-ready output for pasting into the team Roses & Thorns Confluence page. This is a composition skill -- it uses the existing /slack skill for search, then applies AI categorization and rewriting. Output: Formatted text displayed in the terminal. You copy/paste your section into the shared team Live Doc on Confluence. Prerequisites Slack credentials configured ( SLACK_TOKEN , SLACK_COOKIE in ~/.fe-skills/.env ). Run /slack-setup if not done. Dependencies installed: cd .claude/skills/slack && uv sync Pipeline Step 1: Parse arguments Extract the --days N flag from user input. Default to 14 days if not specified. Calculate the search start date: START_DATE = today minus N days, formatted as YYYY-MM-DD For example, if today is 2026-03-16 and --days 14 , then START_DATE = 2026-03-02 . Common invocations: /roses-and-thorns -- last 14 days (default) /roses-and-thorns --days 21 -- last 3 weeks (useful when the biweekly cadence slips) /roses-and-thorns --days 7 -- just last week Step 2: Identify the current user Look up the authenticated Slack user so the search is scoped to whoever is running the skill: uv run --project .claude/skills/slack python .claude/skills/slack/scripts/users.py whoami --pretty Extract the user field from the response (e.g., astevenson ). Store as USERNAME . Step 3: Search Slack for user's posts Use the Slack skill's search.py to find all posts by the user in the time period: uv run --project .claude/skills/slack python .claude/skills/slack/scripts/search.py --query "from:@USERNAME after:YYYY-MM-DD" --count 100 -- sort timestamp --sort-dir desc Replace USERNAME with the value from Step 2 and YYYY-MM-DD with the calculated START_DATE . Pagination: Check messages.paging.pages in the response. If there are multiple pages, fetch each additional page: uv run --project .claude/skills/slack python .claude/skills/slack/scripts/search.py --query "from:@USERNAME after:YYYY-MM-DD" --count 100 -- sort timestamp --sort-dir desc --page N For each message, extract: text -- the message content channel.name -- which channel it was posted in ts -- timestamp (convert to human-readable date for display) permalink -- Slack link to the message reply_count -- whether the message started a thread Thread handling: If a message has reply_count > 0 , note it but do NOT fetch the full thread by default. The top-level post captures the user's contribution. Thread replies are often responses from others and add noise. Deduplication: If the same text appears in multiple channels (cross-posts), keep only the first occurrence by timestamp. Filtering: Skip purely administrative messages: Standup bot responses or form submissions Emoji-only messages Link-only shares without any accompanying text/context Automated workflow messages Step 4: Categorize posts Using your own judgment, assign each post to one or more of these categories: Category What belongs here Highlights Customer wins, progress, positive engagement, successful deliveries, commitments fulfilled, hackweek demos, team achievements Lowlights Customer frustrations, outages, near-misses, time sinks, churn signals, blocked work, things that went wrong Learnings Process insights, tool discoveries, productivity improvements, "TIL" moments, workflow optimizations Risks Renewal concerns, escalation patterns, items with forward-looking negative impact, things that could go wrong Consolidation rules (CRITICAL — this is not a post-by-post listing): ACCURACY IS NON-NEGOTIABLE: NEVER fabricate connections between unrelated posts. Two posts mentioning the same customer does NOT mean they are about the same topic. NEVER attribute actions to the user that they didn't do. Sharing news about a feature ≠ shipping that feature. Discussing an issue ≠ fixing it. Read each post carefully for what the user ACTUALLY DID. NEVER merge posts unless they are genuinely about the exact same specific topic (e.g., multiple updates on the same parquet backfill). Same customer is NOT enough — same customer + same issue is required. When in doubt, keep items separate rather than merge incorrectly. Consolidation approach: Only merge posts that are clearly sequential updates on the SAME specific issue or task. Target 3-5 items per category , but categories DO NOT need to be equal. If the period had more lowlights than highlights, show more lowlights. Reflect reality, not a balanced template. Do NOT bias toward optimism. SEs often have more lowlights and risks than highlights — that's the nature of post-sales work. If the raw posts skew negative, the output should too. Achieve the target count by PRIORITIZING the most impactful items, not by aggressively merging unrelated ones. Drop low-signal items (routine updates, minor admin, chat banter, link-only shares) to reduce count. A consolidated item CAN appear in multiple categories (e.g., "Both my lowlights are also risks"). Other Activities should have 1-3 items max — only genuinely notable things (hackweek, events, initiatives). What the user DID vs what the user SAW (this is the #1 failure mode — get it right): Claude has a strong tendency to use heroic/inflated verbs. Fight this actively. Default to MODEST verbs unless the post CLEARLY shows the user personally did the work. What the post shows WRONG verb RIGHT verb Shared a product release in channels "Shipped", "Launched", "Delivered" "Shared", "Announced", "Highlighted to customers" Coordinated getting something built "Built", "Drove", "Engineered" "Coordinated", "Pushed for", "Got X built" Posted about a customer issue "Fixed", "Resolved", "Solved" "Tracked", "Investigated", "Raised" Escalated to eng "Drove resolution", "Led the fix" "Escalated", "Raised priority" Shouted someone out "Delivered the work" "Recognised X's work" Discussed in a meeting "Drove alignment", "Led the strategy" "Discussed", "Proposed", "Raised" When in doubt, use the less heroic verb. The user can always upgrade it during review. IMPORTANT -- User review checkpoint: Present the categorized list to the user for review BEFORE proceeding to Step 5. Format as: