Stellar Project Dossier

Assemble a structured due-diligence dossier on ONE Stellar project: what it is, its category/tags/links, its Stellar Community Fund (SCF) funding history, what the ecosystem has written and said about it, and the projects most like it. This skill runs entirely on LumenLoop's free, read-only Stellar ecosystem MCP (https://mcp.lumenloop.com) — every fact below comes from a public directory, the public SCF record, and LumenLoop's editorial/AV/research library.

When to use this skill

• The user names a Stellar project and asks to "vet", "profile", "research", "do diligence on", or "summarize" it.
• You need a project's funding track record (which SCF rounds, what award type).
• You want to know what articles, talks, podcasts, or events discuss a team — with citable source urls.
• You need a list of comparable / competing projects to position one team.
• You are prepping a one-pager or background brief on a Stellar builder.

Related skills

• New to the MCP? Connect and learn the toolset first → ../lumenloop-mcp-connect/SKILL.md
• Mapping a whole sector instead of one project → ../stellar-ecosystem-scout/SKILL.md
• Positioning an SCF idea against prior submissions → ../scf-submission-radar/SKILL.md
• Choosing which project to depend on (wallet / oracle / anchor / DEX) → ../stellar-integration-finder/SKILL.md
• Building on Stellar (Soroban contracts, JS stellar-sdk + Freighter + Stellar Wallets Kit, classic assets + trustlines + SAC, Stellar RPC / Horizon data, SEP/CAP standards, the Reflector oracle): see the Stellar developer docs at https://developers.stellar.org.

The dossier pipeline (run in order)

Each step is independent of the others except for the slug — once you have it from step 1, steps 2–6 can be issued together.

1. Resolve the slug

The user gives a name; the tools key off a slug. Resolve it first.

search_directory(query="lobstr")

search_directory is keyword/ILIKE over title, description, and other_names. Keep the term SHORT — one or two words. A long natural-language phrase ("the lobstr non-custodial wallet app") matches nothing (a zero-hit response includes a recovery hint). It returns one JSON object shaped like:

{ count: 1, projects: [
  { slug: "lobstr", title: "LOBSTR", category: "Applications",
    tags: ["Wallet Infra", "Custody"], website: "https://lobstr.co",
    description: "Non-custodial Stellar wallet ..." } ] }

Pick the row whose title/website matches the user's intent and lock its slug. If the user already supplied a slug, skip to step 2. If the name is ambiguous or returns nothing, try a shorter or alternate term, then confirm the match with the user before spending the rest of the pipeline on it.

2. Pull the profile

get_project(slug="lobstr")

The default response is the FULL row — long-form description, based_in / operating_region, twitter_handles, website, all tags, other_names, and more. That is what a dossier wants. If you only need a quick header (title, short description, category, tags, links), pass compact: true for a slim ~500-char shape and save tokens:

get_project(slug="lobstr", compact=true)

Record: category (one of the five from get_categories — Applications, Developer Tooling, Education & Community, Financial Protocols, Infrastructure & Services), project tags (vocabulary from get_project_tags_vocabulary, e.g. "Wallet Infra", "Cross-Border Payments", "SDK"), region, and every link.

3. Funding history (SCF)

get_scf_submissions(slug="lobstr")

Returns the project's Stellar Community Fund submissions. For each, note the round and award_type — this is the funding track record. If the slug returns nothing but you suspect SCF history under a slightly different name, fall back to the fuzzy name lookup:

get_scf_submissions(name="LOBSTR")

An empty result is itself a finding: "no SCF submissions on record." State it plainly rather than omitting the section.

4. What's written and said about it

find_content_about_project(slug="lobstr", limit=10)

Returns content across domains (articles, AV/video, events, research) that DISCUSSES this project. Each item carries title, url, domain, publishing_date, and an AI-generated summary / long_summary. These are LumenLoop's SUMMARIES, not full article text — cite the url for every claim and never quote the summary as if it were the source's exact words.

Ranking here is recency-aware, so recent coverage (a security incident or exploit, an audit, a launch) surfaces alongside the canonical/older items — for a due-diligence dossier, scan explicitly for any incident or audit and date it (publishing_date, flagged as possibly outdated if old). The corpus is the source of record for these ecosystem facts and is dated, so cite the date; reach for live web only for facts it can't hold (e.g. a current on-chain TVL or an upstream package version), not for "has this project had an incident".

If you want a theme-scoped sweep instead (e.g. only what's been said about the project's cross-border payments work), use semantic content search and narrow the types:

search_content_semantic(query="LOBSTR cross-border payments",
                         types=["articles","av"], limit=8)

5. Direct quotes from talks and podcasts

find_av_passages(query="LOBSTR wallet")

This matches per-chunk inside AV transcripts, so it surfaces the actual passage where something was SAID — ideal for a pull-quote in the dossier. Optionally bound it with date_start / date_end to stay recent. Attribute each quote to its source item and url.

6. Comparables / competitors

find_similar_projects_semantic(slug="lobstr", limit=8)

This is semantic project→project similarity (embeddings), which beats tag/category matching for finding real competitors and neighbors. Use the result as the "comparables" section. Spot-check one or two with get_project(slug, compact=true) to confirm they are genuine peers.

Do NOT use get_related_projects for this — it maps a piece of CONTENT to the projects it mentions, not project→project. For project similarity always use find_similar_projects_semantic.

7. (Optional) Ecosystem connections via a key content item

When a step-4 item is especially relevant (a roundup, a partnership announcement), expand who ELSE it names to map the project's neighborhood. Take that item's id and its content type and call:

get_related_projects(content_id=4217, content_type="article")

content_type is one of article | av | event | research (plural/alias accepted); the source must be published. This returns the other projects that item mentions — useful for "appears alongside" context. Then get_project(slug, compact=true) any newly surfaced slug worth a line.

Worked example: the "lobstr" dossier

1. search_directory(query="lobstr") → one strong hit, slug lobstr, category Applications, tags include "Wallet Infra".
2. get_project(slug="lobstr") → full profile: non-custodial Stellar wallet, website, twitter handle, region, long description.
3. get_scf_submissions(slug="lobstr") → any SCF rounds + award_type (or "none on record").
4. find_content_about_project(slug="lobstr", limit=10) → articles and talks mentioning LOBSTR, each with a url to cite.
5. find_av_passages(query="LOBSTR wallet") → a quotable passage from a talk.
6. find_similar_projects_semantic(slug="lobstr", limit=8) → comparable wallets/apps to list as the competitive set.
7. (Optional) pick the most relevant article from step 4 → get_related_projects(content_id=<id>, content_type="article") → who else it names.

Assemble the results into the dossier structure below.

Output: the dossier

Produce a single structured brief with these sections:

• Overview — one-paragraph what-it-is (from step 2).
• Profile — category, project tags, region, all links (website, twitter).
• SCF / funding — each submission's round + award_type, or "none on record".
• What people say — 3–6 bullets, each a one-sentence takeaway + the source url; include one attributed pull-quote from step 5 if available.
• Comparables — 3–8 similar projects (slug + one-line why), from step 6.
• Open questions — gaps the public data did not answer (e.g. no recent content, no funding record, unclear traction) to flag for follow-up.

Full section-by-section template, with a filled example and citation rules → reference/dossier-template.md.

Gotchas

• search_directory is ILIKE keyword search — SHORT terms only; long phrases return 0. For topic/concept discovery use the semantic tools.
• The content tools return AI SUMMARIES + a source url, not full article text. Cite the url; never present a summary as a verbatim quote.
• get_related_projects is content→projects (needs content_id + content_type), NOT project→project. Use find_similar_projects_semantic for comparables.
• get_project defaults to a large full row; pass compact: true when you only need the header.
• Empty results are findings — state "none on record" rather than dropping a section.
• If a tool stops appearing after a server update, reconnect the connector (see ../lumenloop-mcp-connect/SKILL.md).

Further reading

• LumenLoop MCP overview and full tool catalog → ../lumenloop-mcp-connect/SKILL.md
• Building on Stellar (Soroban, stellar-sdk, assets, data, standards, Reflector) → https://developers.stellar.org