Stellar Ecosystem Digest

Produce a dated, cited digest of recent Stellar ecosystem activity — grouped into News / Talks & Video / Events / Research with one-sentence takeaways and source URLs, plus an "entities to watch" list. It runs entirely on LumenLoop's free, read-only Stellar ecosystem MCP at https://mcp.lumenloop.com, which indexes the public project directory plus editorial news, audio-video (talks/podcasts), events, and a curated research library.

When to use this skill

• The user asks "what's new / recent / happening" in an area of Stellar (DeFi, RWA, wallets, anchors, payments, Soroban) and wants it summarized.
• They want a newsletter-style roundup or weekly/monthly digest of a theme.
• They name a project, person, organization, or token and want recent coverage of it.
• They want a dated catch-up brief with citations they can click through.
• They ask "what should I be watching" in a sector and want entities to track.

Related skills

• LumenLoop MCP setup & the full tool catalog → ../lumenloop-mcp-connect/SKILL.md
• Map a whole sector into a landscape (not just recent items) → ../stellar-ecosystem-scout/SKILL.md
• Go deep on one project surfaced by the digest → ../stellar-project-dossier/SKILL.md

For BUILDING on anything the digest surfaces (Soroban contracts, the JS stellar-sdk / Freighter / Stellar Wallets Kit, classic assets + trustlines + SAC, Stellar RPC / Horizon data, SEP/CAP standards, the Reflector oracle), point the user at the official docs: https://developers.stellar.org.

The pipeline

A digest is built in six passes. Run only the passes the request needs — a theme digest leans on passes 1–2 and 4–5; an entity digest leads with pass 3.

Every content tool returns LumenLoop's AI-generated summary plus the source url and publishing_date — cite the url and date; the summary is not full text.

The corpus is the source of record for ecosystem freshness — it is dated and updated continuously, and semantic ranking is recency-aware — so build the digest from it and cite each item's date. You do not need live web for "what's new" in the ecosystem; reserve web for facts the corpus can't hold (e.g. an upstream package/registry version pin), not for recent ecosystem activity.

Choosing the date window

Resolve relative phrases against today's date before calling. Pass ISO date_start / date_end (YYYY-MM-DD). Common windows:

Pass 1 — newest items per collection

Lead with the raw recency feed so nothing recent is missed, then filter by theme.

list_documents(collection="articles", sort="publishing_date", order="DESC", limit=15)
list_documents(collection="videos",   sort="publishing_date", order="DESC", limit=10)
list_documents(collection="research",  sort="publishing_date", order="DESC", limit=10)
list_documents(collection="events",   period="upcoming", limit=10)

• collection ∈ articles | events | videos | jobs | research (videos and av map to the same table; av is accepted as an alias for videos).
• period applies to events only: all | upcoming | past.
• Use search to narrow by a title/URL substring when you already have a keyword.
• Returns rows with id, title, url, publishing_date, source/domain, and a summary — keep the id so you can get_document later.

Pass 2 — theme filter (semantic, dated)

Narrow the firehose to the requested theme. Semantic search beats keyword here because phrasing varies across outlets.

search_content_semantic(
  query="Soroban DeFi lending and liquidity on Stellar",
  types=["articles", "av", "research"],
  date_start="2026-03-08",
  date_end="2026-06-08",
  limit=12,
)

• types ⊆ articles | av | events | research. Narrow the set for speed and relevance — default is broad. limit defaults to 5 per content type, so raise it (as above) for a digest sweep.
• Returns per-type arrays of {id, title, url, domain, publishing_date, summary, slug, similarity}; pass response_format="detailed" to add each hit's long_summary. Sort each group by publishing_date DESC for the digest; use similarity to drop weak matches (weak hits may also carry a _weak_match flag plus a _hint).
• For a salient article you want to anchor the digest on, map it to the projects it mentions with get_related_projects(content_id=<id>, content_type="article") — that feeds the "entities to watch" list. content_type ∈ article | av | event | research.

Pass 3 — entity focus

When the digest is about a named project / person / org / token, lead here.

find_content_by_entity(
  entity="Reflector",
  entity_type="project",
  content_type="articles",
  date_start="2026-03-08",
  date_end="2026-06-08",
  min_confidence=0.6,
)

• entity is required. entity_type ∈ project | person | organization | token.
• content_type ∈ articles | av | events | scf_submissions.
• min_confidence defaults to 0.5; raise to 0.6–0.7 to cut loose mentions.
• Entities are LLM-extracted, so try other_names if a brand has aliases.

Pass 4 — quotes from talks

To quote what someone actually said (not a summary), search inside AV transcripts. This returns per-chunk passages, ideal for a pull-quote in the "Talks & Video" group.

find_av_passages(
  query="real-world asset tokenization on Stellar",
  date_start="2026-03-08",
  date_end="2026-06-08",
  limit=6,
)

Each hit carries the matched passage, the parent video's title/url, and date — quote the passage and cite the url.

Pass 5 — research library

LumenLoop's curated research library is its own collection. Pull it two ways:

list_documents(collection="research", sort="publishing_date", order="DESC", limit=8)
# or, theme-scoped:
search_content_semantic(query="<theme>", types=["research"], date_start=..., date_end=...)

Pass 6 — expand a flagged item

When one item deserves a paragraph instead of a line, fetch its full record:

get_document(collection="articles", id=<id>)

collection ∈ articles | events | videos | av | jobs | research (videos/av share a table). Use this for the lead story or a research piece you want to brief in depth — but the body is still a structured record, so cite the url.

Worked example — "What's new in Stellar RWA in the last 30 days?"

Today is 2026-06-08, so the window is 2026-05-09 → 2026-06-08.

1. Recency feed. list_documents(collection="articles", sort="publishing_date", order="DESC", limit=15) and the same for videos and research — skim for anything RWA-flavored, keep the ids.
2. Theme filter. search_content_semantic(query="real-world asset tokenization on Stellar", types=["articles","av","research"], date_start="2026-05-09", date_end="2026-06-08", limit=12). Sort each per-type array by date; drop low similarity.
3. Entity sweep. For each issuer that recurs (e.g. an RWA platform), run find_content_by_entity(entity="<name>", entity_type="project", date_start="2026-05-09", date_end="2026-06-08", min_confidence=0.6).
4. Pull a quote. find_av_passages(query="tokenizing treasuries on Stellar", date_start="2026-05-09", date_end="2026-06-08", limit=6) for a talk soundbite.
5. Research. search_content_semantic(query="real-world assets on Stellar", types=["research"], date_start="2026-05-09", date_end="2026-06-08").
6. Entities to watch. Take the lead article's id, call get_related_projects(content_id=<id>, content_type="article"), then get_project(slug, compact=true) on each to get a one-line tag/category gloss.

Note: search_directory is keyword/ILIKE over title/description/other_names — short terms like RWA or tokenization work, but a long natural-language phrase returns 0 rows. For recent activity always reach for the semantic + list_documents tools above, not search_directory.

Output shape

Assemble a dated brief grouped by content type. Use the template in reference/digest-template.md. Each line is one takeaway + a cited link:

# Stellar RWA — digest for 2026-06-08 (last 30 days)

## News
- Issuer X opens tokenized treasuries to retail anchors. — example.com/x (2026-05-22)

## Talks & Video
- "We settle the asset on-ledger, KYC stays off-ledger." — talk title, example.com/v (2026-05-30)

## Events
- RWA on Stellar workshop, virtual, 2026-06-20. — example.com/e (upcoming)

## Research
- LumenLoop: tokenized-RWA adoption trends across anchors. — research url (2026-05-15)

## Entities to watch
- issuer-x — Financial Protocols · tags: RWA, Anchor — example.com

Quality rules

• Always cite. Every line carries the source url; summaries are AI-generated, not the article text — never present them as direct quotes (only find_av_passages gives verbatim passages).
• Always date. Put publishing_date on each item and the window in the title; sort newest-first within each group.
• De-dupe across passes. The same item appears via recency, theme, and entity passes — key on id/url and keep one entry.
• Stay current. If a tool stops appearing after a server update, reconnect the LumenLoop connector (see ../lumenloop-mcp-connect/SKILL.md).

Pointers

• Build on what you find: https://developers.stellar.org
• Connector & full tool catalog: ../lumenloop-mcp-connect/SKILL.md
• Output template: reference/digest-template.md