Stellar Content Auditor

Paste in a draft — a blog post, talking points, a tweet/thread, a launch announcement — and this skill checks and enriches it against LumenLoop's free, read-only Stellar ecosystem MCP (https://mcp.lumenloop.com). It resolves every project you mention to its canonical directory entry (correct name, slug, X/@ handle, website), backs up factual claims with cited indexed content, and flags anything wrong, missing, or unsupported — so what you publish is accurate and well-linked.

When to use this skill

• The user shares draft copy about Stellar and wants it checked before publishing.
• "What's the right @handle / official name for these projects?"
• "Fact-check this post" / "add citations and links" / "what should I also mention?"
• Verifying a launch/announcement names projects and partners correctly.
• Catching misspelled or non-existent projects and wrong X handles in copy.

Related skills

• Connect the MCP + the tool map → ../lumenloop-mcp-connect/SKILL.md
• Deep-dive one project you mention → ../stellar-project-dossier/SKILL.md
• Find recent content on a topic to cite → ../stellar-ecosystem-digest/SKILL.md
• Map the wider space around your draft → ../stellar-ecosystem-scout/SKILL.md
• Verify an SCF funding claim → ../scf-submission-radar/SKILL.md

The audit workflow

1. Extract the entities from the draft

Read the draft and list everything that names the ecosystem: projects, people, organizations, tokens, plus any @handles and factual claims (funding, metrics, "first to…", partnerships). Work from the user's text only.

2. Resolve each project to its canonical entry

For every project mentioned, find its directory record:

search_directory(query="<short project name>", limit=5)
   → {count, projects: [{slug, title, description, category, tags, website}]}
get_project(slug="<slug>", compact=true)
   → { slug, title, short description, category, tags, twitter_handles, website, other_names }

• search_directory is keyword/ILIKE — use the short name ("aquarius", not "the Aquarius AMM protocol"). If the draft's spelling is off, try other_names-style variants or a shorter token.
• If search_directory returns 0 results, the project may not be in the directory (or is misspelled): flag it — do not invent a slug or handle.
• get_project(compact=true) returns the canonical title, twitter_handles, and website — this is your source of truth for names and @mentions.

3. Check names and @handles

Compare each as-written name/@handle to the canonical title / twitter_handles:

• Wrong or missing handle → propose the canonical one.
• Wrong capitalization / spelling of the name → propose the canonical title.
• Handle in the draft but project not in the directory → flag as unverified.

4. Fact-check claims and attach citations

For each factual claim, look for supporting indexed content:

find_content_by_entity(entity="<name>", entity_type="project", limit=10)
search_content_semantic(query="<the claim, in words>", types=["articles","av","research"])
find_content_about_project(slug="<slug>", limit=8)

Each returns LumenLoop's AI summaries + the source url. Cite the url.

• Claim supported by indexed content → attach the citation.
• Claim with no supporting content → mark unverified (do not fabricate a source).
• Need a direct quote from a talk/podcast → find_av_passages(query="<topic>").
• Claim about SCF funding → get_scf_submissions(slug="<slug>") to confirm round/award.

5. Suggest links and mentions

Strengthen the draft:

• find_similar_projects_semantic(slug="<slug>") → adjacent projects worth mentioning.
• search_content_semantic(query="<theme>") → recent articles/research to link.

6. Produce the audit

Return the draft annotated with three things: a mentions table (as-written → canonical name / slug / @handle / website → status), a claims table (claim → citation url → verdict), and suggestions (projects/content to add). Use the layout in reference/audit-report-template.md.

Worked example

Draft says: "Big fan of Aquarious — the AMM on Stellar (@aquarius) just crossed $1B."

1. search_directory(query="aquarius", limit=5) → slug aquarius, title "Aquarius".
2. get_project(slug="aquarius", compact=true) → canonical title Aquarius (draft misspells "Aquarious"); twitter_handles gives the real handle (verify @aquarius matches — correct it if not).
3. find_content_by_entity(entity="Aquarius", entity_type="project") + search_content_semantic(query="Aquarius AMM total value locked $1B") → look for a source for the "$1B" claim; attach the url if found, else mark unverified.
4. find_similar_projects_semantic(slug="aquarius") → suggest a related AMM/DEX to mention.

Output: "Aquarious → Aquarius (slug aquarius); confirm @handle; the $1B claim is unverified — no indexed source. Consider also linking [related project]."

Gotchas

• search_directory is keyword, not semantic — short terms only; long phrases return 0. For fuzzy spellings, try shorter tokens before concluding "not found".
• Never invent handles or citations. If a project isn't in the directory or a claim has no supporting content, say so explicitly.
• Content tools return summaries + a url, not full text — cite the url; don't present a summary as a verbatim source quote.
• twitter_handles from get_project(compact=true) is the canonical @handle — use it to correct @mentions; it may differ from the draft.
• Treat directory metrics/claims as directional — confirm time-sensitive numbers against a dated source before vouching for them.

Pointers

• LumenLoop MCP server: https://mcp.lumenloop.com
• Building on Stellar: https://developers.stellar.org