Gemini CLI: Claude Cowork Content Radar Agent v2

This workflow uses Claude Cowork as the orchestration layer to run a content radar system that scans 30+ RSS feeds and Reddit communities, scores each content idea against a brand profile on a 0-10 scale, checks the existing content calendar for conflicts, and drafts a 5-minute morning brief ranking what to publish next. The agentic reasoning step is the scoring mechanism: Claude evaluates each trending topic against three weighted dimensions — brand relevance, audience demand signal, and production feasibility — and only surfaces ideas that cross a configurable threshold (default 7/10). This is distinct from basic RSS aggregation where you manually scan headlines. The system includes a Sunday recalibration loop that reads notes from published posts and retunes the scoring profile for the coming week. Teams using this workflow report cutting content planning time from 10-14 hours per week to under 3 hours while maintaining or improving post performance.

BUSINESS PROBLEM

Content teams spend 10-14 hours per week manually scanning feeds, evaluating ideas against brand goals, and checking calendars before writing a single post. A 2025 Content Marketing Institute survey found that 63% of B2B marketers say producing engaging content is their biggest challenge, and 54% say they struggle with content ideation specifically. The typical routine involves opening 20+ browser tabs, scanning headlines, copying links into a spreadsheet, and then having a 60-minute editorial meeting to decide what to write. For a team of three content marketers at $75,000 average salary each, that weekly planning overhead costs approximately $1,200 per week in labor. Worse, editorial decisions are driven by whoever has the strongest opinion in the meeting, not by data. Good ideas that the loudest person did not notice get skipped. A systematic trend-to-publication pipeline eliminates both the time cost and the recency bias that plagues manual editorial planning.

WHO BENEFITS

Content marketing teams at SaaS companies (10-50 person marketing departments) publishing 4-8 posts per week who need to balance trending topics with brand strategy. Solo newsletter writers or ghostwriters managing 2-3 client publications who must track multiple industry trends simultaneously without editorial support. In-house editorial leads at media companies who oversee 5+ writers and need a daily publishing brief that replaces the morning editorial meeting. Each profile faces a version of the same problem: trend discovery is scattered, scoring is subjective, and calendar conflicts surface too late.

HOW IT WORKS

1. Feed Collection (Background, Continuous). Claude Cowork maintains a list of 30+ RSS feeds categorized by source type (industry news, competitor blogs, academic journals) and 10 Reddit subreddits relevant to the brand's vertical. The system polls these every 4 hours via the RSS API and Pushshift Reddit API. Output: raw feed JSON with title, URL, publish date, and source category.
2. Deduplication and Prioritization. Claude Cowork deduplicates items that appear across multiple feeds and assigns a priority tier based on source authority. Breaking news from primary sources gets Tier 1 priority. Niche blog posts get Tier 3. Output: a deduplicated queue of 50-200 items per cycle.
3. AI Scoring Against Brand Profile. Claude evaluates each item against the brand's stated content pillars, audience personas, and editorial guidelines stored in a configuration file. It assigns a score from 0-10 based on three criteria: brand relevance (40% weight), audience demand signal based on Reddit upvote velocity or RSS recency (35%), and production feasibility considering available resources and expertise (25%). Items scoring 7 or above pass to the next stage. This is the primary reasoning step.
4. Calendar Conflict Check. The system queries the Google Calendar API for scheduled posts in the next 14 days. It checks for topic overlap, format duplication (two listicles in one week), and pillar imbalance (all posts covering the same content pillar). Conflicting ideas are either deprioritized or returned with a modification suggestion.
5. Morning Brief Generation. Claude Cowork drafts a 5-minute morning brief in plain text. The brief lists the top 3-5 publish-ready ideas with their scores, a one-sentence rationale, and a recommended post format. Output: a markdown-free text brief pushed to Slack or email via API.
6. Human Review and Selection. The content team reviews the morning brief and selects the idea(s) to execute. This takes approximately 5 minutes. The selected idea is automatically added to the Google Calendar via API. Checkpoint.
7. Sunday Recalibration Loop. Every Sunday, Claude Cowork reads the performance data from the week's published posts (engagement, clicks, shares) and reads any editorial notes the team left. It adjusts the three scoring weights by up to 2 points per dimension and rewrites the brand profile if the notes indicate a strategic shift.

TOOL INTEGRATION

Claude Cowork (via Gemini CLI desktop or CLI): Central orchestration and LLM agent. The scoring profile is stored as a plain text config file. Gotcha: Claude Cowork does not persist state between sessions by default. You must configure a persistent memory directory using the --memory-dir flag and point it to a git-tracked folder so the Sunday recalibration loop has access to historical scores. Reddit API (via Pushshift or official API): For subreddit trend scanning. Free tier allows 60 requests per minute with OAuth. Gotcha: Reddit's official API rate limits tightened in 2023 after the API pricing change. Use Pushshift as a fallback for historical data (free, no auth needed) and only use the live API for current-48-hour data. RSS Feed Reader Integration: Use an RSS-to-JSON converter API like rss2json.com (free tier: 1000 requests/month) or self-host a FreshRSS instance. Gotcha: Many RSS feeds have moved to Atom format. The rss2json converter handles both RSS and Atom, but the free tier does not support authenticated feeds behind login walls. Google Calendar API: For calendar conflict checking and post scheduling. Requires OAuth scope https://www.googleapis.com/auth/calendar. Gotcha: Google Calendar API has a 10,000 requests per day quota on the free tier. Each scheduling check reads 14 days of events. For teams checking 3-4 times daily, this uses 180-240 requests. Add exponential backoff retry logic to handle 429 errors.

ROI METRICS

1. Content planning time: 10-14 hrs/week → 2-3 hrs/week. Measurable in week 1 by timing the morning brief review. 2. Ideas surfaced per week: 5-10 manual → 30-50 scored and filtered by the radar system. 3. Calendar conflict detection: manual checks miss 2-3 conflicts per month → zero conflicts with automated calendar scan. 4. Cost per planning cycle at $60/hr blended team rate: $600-$840/week → $15-$30/week in API and compute costs. 5. Post performance improvement: teams report 20-40% increase in click-through rates when publishing data-selected topics vs. meeting-selected topics (Source: Content Marketing Institute, 2025).

CAVEATS

1. Scoring drift: The Sunday recalibration adjusts weights based on a single week of data, which can overcorrect for anomalous performance. A viral post about an off-brand topic could skew the scoring profile toward that topic. Implement a two-week rolling average for weight adjustments. 2. Reddit data quality: Subreddit trends can be astroturfed. The system has no detection for coordinated upvote campaigns. Items that score high due to artificial engagement will be recommended. 3. RSS feed breakage: Feeds go stale without notice. An RSS feed that stops updating silently reduces the signal pool. The system does not currently alert on stale feeds. Monitor feed freshness manually once per month. 4. This workflow does not write the actual content. It produces recommendations and briefs only. Writing, editing, and publishing remain human tasks.