💡 Daily feed summarizer agent

==tl;dr: A Claude Code skill + scheduled routine that reads Twitter (and ideally Reddit) once a day and produces a digest, so the slop never has to be consumed directly.==

I’ve been thinking about implementing a new skill and a Claude code routine that basically every day reads my Twitter feed and summarizes it so that I don’t have to consume the slop myself. Also same for Reddit, but I’m not sure if the Reddit API allows it. So for X, I know that there’s a CLI tool that can be used that was recently released, so I guess that can be used maybe. But for Reddit, I’m not sure.

Why?

• Consuming Twitter/Reddit directly is net-negative — high volume, low signal, actively engineered for engagement.
• A daily digest keeps the signal (interesting threads, relevant posts from followed accounts) while dropping the slop (outrage bait, recommendation algorithm drift).
• Fits the AI-native PKM vision: the agent consumes feeds on the human’s behalf, surfaces what matters, stores rest.
• Parallels the existing Feediary idea (scrape-based aggregator) but routes through the Claude Code agent loop instead of a standalone app.

What?

• Skill: feed-digest (or similar) — fetches a feed source, filters/summarizes, writes a dated digest note into the vault.
• Routine: scheduled daily (e.g., morning) via the existing Claude Code routine mechanism.
• Output: a single daily digest note per source (0_Inbox/feeds/twitter-YYYY-MM-DD.md, reddit-YYYY-MM-DD.md) or a merged feed. Processed by process-inbox like any other inbox item — the agent can link interesting items into related pages, drop the rest.

How?

• Twitter (X): a CLI tool for X was recently released — likely usable as the data source. Needs investigation: which tool, auth model, rate limits, whether the home timeline is accessible or only public search.
• Reddit: unclear whether the Reddit API still permits this post-2023-API-changes. Options:
  • Official Reddit API (needs auth, subject to rate limits + access tier).
  • Personal RSS feeds (Reddit exposes .rss per-subreddit / per-user — cheaper path, no API key).
  • Third-party scraper — fragile, TOS risk.
• Summarization: Claude via the existing Claude Code agent. Each item scored (interesting vs. slop), then a digest produced with the interesting items summarized and the slop discarded.
• Inbox integration: digest is routed like any other inbox note. If an item warrants its own page (a thread worth keeping), process-inbox promotes it.

Open questions

• What’s the CLI tool for X/Twitter?
• Does the Reddit API still allow personal-feed reads in 2026?
• Per-source digests or merged “morning digest” across all feeds?
• Where do digests live long-term — 0_Inbox/feeds/ (processed and deleted) or 3_Resources/Feeds/<source>/<date> (kept as archive)?

Related

• 👨‍💻 Apps — Feediary (same problem, standalone-app framing).
• OpenClaw for PKM — “Proactive Task Execution” + “Controlled Feedback Loop” are the conceptual parents (daily digest, not random interruptions).
• 2026-01-10-Unified social feed with topic curation — earlier private memo raising the same frustration (Reddit + X aggregation, topic curation, API costs blocking it). This idea is the agent-loop-era answer.
• AGENTS.md — skill + routine contract.
• SKILL — would route the daily digests.
• SKILL — precedent for a skill that processes raw external input into wiki-ready form (Phase 0.5 action-item detection, routing, graph enrichment).
• Open-Source Model-Agnostic AI Platform — this skill would be a concrete routine on that kind of open platform.
• Agent-Native Software Architecture — the architectural pattern this skill sits inside: a skill (function) called by a sub-agent, using MCP/CLI connectors.