Aggregator
SkillFiles & storageDaily fetch from a fixed public allowlist; score against the user's interests file; cluster into themes; push the digest to the configured channel
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Aggregator skill
What this skill tells your AI
The instructions your AI receives, as published by gebruder/wirken in preset/zirkel/skills/aggregator/SKILL.md and read by ahel’s review.
Fires once per day on a cron the operator configures. Performs a strict-pipeline run over the source allowlist:
- Fetch each source in
sources.tomlvia the declared method (RSS / API / rate-limited scrape). The HTTP wrapper enforces the egress allowlist. - Normalize fetched items into candidate records:
title,source,date,url,body_excerpt,citation_json. - Skip if the candidate's content hash is already in the seen set.
- Score the candidate's relevance against the user's interests file (
~/.wirken/zirkel/interests.toml). Output: a float in[0, 1]plus a one-linewhy_surfacedrationale that names the matched interest. - If the score crosses the keep threshold and no skip pattern matches, write the candidate to the local SQLite store. Otherwise, log the skip with reason.
- After all sources fetched, cluster the run's kept candidates into emergent themes using a small local embedding model.
- Render the digest as plain prose with footnote-ready citations on every item. Push to the configured channel adapter.
Aggregator does not synthesize claims about the world. The digest reads as "N items kept, M skipped. Today's themes: [theme A: items 1, 4; theme B: items 2, 3, 7]. Each item links to its source with a one-line why-surfaced explanation." No "today the regulators are saying X" prose.
This skill is disable-model-invocation: true. Reach it via /aggregator (manual run) or via the operator's cron that calls the underlying CLI command directly.
State
~/.wirken/zirkel/zirkel.db— SQLite database holding candidates, seen, themes, skipped_log, runs, and the cached interests snapshot per run.~/.wirken/zirkel/interests.toml— user-editable; reloaded at the start of every run; a hash of the file is recorded in the audit log so changes are visible run-to-run.~/.wirken/zirkel/bodies/— full bodies referenced bycandidates.body_full_pathfor items where the excerpt is too small for retrieval.
The aggregator and librarian share this state; both skills are scoped to the same ~/.wirken/zirkel/ directory by their permissions blocks.
Inputs the operator provides
sources.toml(preset-level, alongsidepreset.toml) — the addressable public set of sources.interests.toml(per-user, in~/.wirken/zirkel/) — concepts, keywords, exclusions, optional source weight tweaks.- Channel adapter pin and target (configured at preset install).
Signals
- GitHub stars
- 171
- Forks
- 5
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
aggregator- Source
- github.com/gebruder/wirken