Lobstr

Tells you if a startup idea is worth building — in 60 seconds. Use when a user wants to evaluate, validate, or score a startup idea; asks "should I build thi...

Security Scan

The skill appears to do what it says (score startup ideas via a hosted API, with optional BYOK mode); its permissions and env vars are proportionate, but you should be aware the idea text is sent to an external hosted API and verify the remaining script logic before installing.

Purpose & Capability

Name/description match requested resources: default mode uses a hosted scoring API (runlobstr.com) and optional ANTHROPIC_API_KEY / EXA_API_KEY are documented and only needed for BYOK/local pipeline. The optional MOLTBOOK_API_KEY is only needed for the --moltbook flag. No unrelated credentials or binaries are requested.

Instruction Scope

SKILL.md explicitly instructs running scripts/lobstr.py and warns that idea text is sent to runlobstr.com by default. This is expected behavior for the stated purpose, but it does mean user-provided ideas (potentially confidential) will be transmitted off-agent. The code includes functions to call Anthropic and Exa APIs for BYOK mode; the SKILL.md and SECURITY.md claim those are only used when both keys are set. I could not fully confirm the main execution path because the script was truncated near the end — verify the main logic ensures BYOK calls only when both keys are set.

Install Mechanism

No install spec is provided (instruction-only with an included script). Nothing is downloaded or written by an installer step. This is the lowest-risk install mechanism.

Credentials

No required environment variables. The optional ANTHROPIC_API_KEY and EXA_API_KEY are reasonable for a BYOK mode that runs local scoring/search, and MOLTBOOK_API_KEY is only relevant to the publish flag. No unrelated secrets or excessive env access is requested.

Persistence & Privilege

always is false and the skill does not request persistent system privileges or write system-wide config. SECURITY.md states the skill does not persist data or write files. Flags allow publishing (--public) or posting (--moltbook) but these are explicit user actions.

Assessment

This skill is coherent with its description: by default it will send the idea text to runlobstr.com and return a private score card; optional environment variables enable a local/BYOK pipeline that will make requests to api.anthropic.com and api.exa.ai. Before installing, consider: (1) Don’t send highly confidential IP or private pitches unless you’re comfortable transmitting them to runlobstr.com (or to Anthropic/Exa if you enable BYOK). (2) Review the script’s main execution path (scripts/lobstr.py) to confirm the BYOK logic only runs local calls when both keys are set (the file provided was truncated near the end). (3) Avoid setting API keys you don’t trust; the skill will use any provided ANTHROPIC_API_KEY / EXA_API_KEY for network calls. (4) If you might publish results, be cautious with --public and --moltbook flags as those explicitly share output externally. If you want higher assurance, inspect the full scripts/lobstr.py end-to-end or run the script in an isolated environment first.

LOBSTR — Startup Idea Scorer

Trigger

Explicit triggers:

• User types /lobstr "their startup idea"
• User types /validate, /scan, or /score followed by an idea

Proactive triggers (ask the user before running):

• User says "should I build this?" or "is this a good idea?"
• User is describing a startup concept they are considering
• User asks for a competitive analysis of a new product idea
• User is brainstorming business ideas and wants structured feedback
• User asks "what do you think of this idea?"
• User mentions a problem they want to solve and is considering a startup

When triggering proactively, say: "Want me to run a LOBSTR scan on that? It'll give you a competitor landscape, pitch score, and EU investor signal in about 60 seconds."

Workflow

Run scripts/lobstr.py with the idea as a single argument:

python3 scripts/lobstr.py "their startup idea"

The script prints the formatted score card to stdout. Show it to the user verbatim — do not reformat or summarize.

If the script errors, surface the error message to the user clearly.

Flags

Flag	Effect
(none)	Private output only — score card to stdout
--public	Also publish to runlobstr.com and show share URL
--moltbook	Also post to m/lobstrscore on Moltbook
--json	Output raw JSON instead of formatted card (for agent-to-agent piping)

Default usage (no flags) makes one outbound call to runlobstr.com/api/score for scoring and returns privately. No data is published or shared.

Agent usage

When another agent calls this skill programmatically, use --json to get structured output:

python3 scripts/lobstr.py "idea" --json

Returns a JSON object with overall_score, dimensions, competitors, grid, verdict, build_it.

What the user gets

• LOBSTR score (0–100) across 6 dimensions: Landscape, Opportunity, Business model, Sharpness, Timing, Reach
• Competitor list with real companies found via live web search
• GRID investor signal — how many EU VCs are active in the space
• Build/don't build verdict — honest, not flattering
• Shareable URL at runlobstr.com (only with --public)

Requirements

No API keys required. LOBSTR uses the free hosted API at runlobstr.com (5 scans/day).

For unlimited scans, set both keys to enable BYOK mode (local pipeline):

export ANTHROPIC_API_KEY=<your-key>
export EXA_API_KEY=<your-key>

Score Card Format (for reference)

🦞 LOBSTR SCAN
"idea here"

LOBSTR SCORE 74/100 [=======---]

L  Landscape   🟢  82/100  one line verdict
O  Opportunity 🟡  71/100  one line verdict
B  Biz model   🟡  65/100  one line verdict
S  Sharpness   🔴  61/100  one line verdict
T  Timing      🟢  88/100  one line verdict
R  Reach       🟢  79/100  one line verdict

VERDICT
Two sentence VC verdict here.

✅ BUILD IT.

Color rules: 🟢 ≥ 70, 🟡 50–69, 🔴 < 50