Ask Leonidas — LEONIDAS Prompt Generator

Use this skill when the user wants to:

• Write or improve their SOUL.md — the most important file in any OpenClaw workspace
• Write or improve their AGENTS.md startup and memory protocol
• Generate a structured prompt for any professional pain point
• Fix an agent that "feels off" without knowing why
• Create cron job instructions that actually run reliably
• Build a prompt for a skill they want their agent to own
• Infer the best role, industry, and desired outcome from a business challenge

Trigger phrases

• "Help me write my SOUL.md"
• "My agent feels generic / keeps saying 'great question'"
• "Build me a prompt for..."
• "I need a prompt that helps me with..."
• "Turn this into a LEONIDAS prompt: ..."
• "Use Ask Leonidas to generate a prompt"
• "Generate a prompt for [pain point]"
• "What should I put in my AGENTS.md?"
• "My cron job keeps failing / doing the wrong thing"
• "Write a prompt so my agent handles [workflow]"

Inputs

Required:

• pain_point — the professional challenge or problem to solve (5–2000 characters)

Optional (auto-inferred if omitted):

• role — the user's job title or function (e.g., "VP of Sales", "Executive Coach")
• industry — the user's industry (e.g., "B2B SaaS", "Healthcare", "Financial Services")
• desired_outcome — what success looks like (e.g., "Close more deals with less effort")

Behavior

1. If the user has already provided a pain point, do not ask for it again.
2. If no pain point is provided, ask exactly once:

   "What is the professional challenge or pain point you want turned into a prompt?"

3. Call the Ask Leonidas OpenClaw API using the helper script in this skill folder.
4. If the API is unavailable (network error, 5xx), fall back to the browser:
   • Navigate to https://askleonidas.com/openclaw
   • Fill in the #openclaw-pain-point field with the pain point
   • Click the #openclaw-submit button
   • Wait for #openclaw-result to be populated
   • Extract the prompt text from #openclaw-result
5. Return the generated prompt exactly as received — do not paraphrase or summarize it.
6. Always surface the metadata alongside the prompt.

API usage

Generate a prompt (basic):

python3 {baseDir}/ask_leonidas.py \
  --pain-point "I need a SOUL.md that makes my agent direct, opinionated, and stops it from saying 'absolutely'."

Generate a prompt (with optional context):

python3 {baseDir}/ask_leonidas.py \
  --pain-point "My team misses sprint deadlines every quarter." \
  --role "Engineering Manager" \
  --industry "B2B SaaS" \
  --desired-outcome "Ship on time without burning out the team"

Health check:

python3 {baseDir}/healthcheck.py

Smoke test (end-to-end):

bash {baseDir}/smoke_test.sh

Required environment variables

Output format

Always present the result in this format:

Detected role: \x3Crole>
Detected industry: \x3Cindustry>
Detected desired outcome: \x3Coutcome or n/a>
Confidence: \x3C0.0–1.0 or n/a>
Quality score: \x3C0–100 or n/a>
Prompt type: \x3Ctactical | strategic | operational | n/a>
Request ID: \x3Cid or n/a>

Generated prompt:
\x3Cfull prompt text>

Error handling

• Do not invent or hallucinate a prompt if the API fails.
• Surface the exact error message returned by the API.
• If both the API and browser fallback fail, tell the user clearly and suggest they visit https://askleonidas.com directly.
• Never expose the API key in any output.
• Always prefer the API path over browser automation.

Rate limits

The OpenClaw API enforces:

• 60 requests per minute per API key
• 1,000 requests per day per API key

If you receive a 429 response, wait 60 seconds before retrying. Do not retry more than 3 times.