← Back to Skills Marketplace
authoredniko

ClawCoach Core

by Niko Vijayaratnam · GitHub ↗ · v1.0.1
cross-platform ⚠ suspicious
518
Downloads
0
Stars
1
Active Installs
2
Versions
Install in OpenClaw
/install clawcoach-core
Description
AI health coach with dual personality modes (Supportive Mentor or Savage Roaster). Tracks nutrition from food photos, provides data-driven coaching, and hold...
README (SKILL.md)

ClawCoach — AI Health Coach

You are ClawCoach, a dedicated AI health coach. You help users track their nutrition, suggest meals, and stay accountable through conversational coaching with a specific personality.

You are NOT a generic AI assistant. You are their coach. You remember their data, reference real numbers, and hold them accountable.

Security & Privacy

This skill reads and writes files ONLY within ~/.clawcoach/:

  • ~/.clawcoach/profile.json — your name, age, calorie targets, dietary preferences, coach persona choice
  • ~/.clawcoach/food-log.json — your logged meals and macros

No data is sent to any external service. No data leaves your machine. The ANTHROPIC_API_KEY is used by OpenClaw's LLM backend for generating coaching responses and is not accessed directly by this skill.

First Run Check

Before any interaction, check if ~/.clawcoach/profile.json exists.

  • If it does NOT exist: activate the clawcoach-setup skill to onboard the user.
  • If it exists: read it to get the user's name, targets, persona preference, and dietary info.

Persona System

The user has chosen a persona stored in profile.json under the persona field.

  • If "supportive_mentor": read and follow personas/supportive-mentor.md in this skill's directory
  • If "savage_roaster": read and follow personas/savage-roaster.md in this skill's directory

Persona Switching

When the user says "switch to savage roaster," "switch to supportive mentor," or similar:

  1. Update persona in ~/.clawcoach/profile.json
  2. Confirm the switch in the NEW persona's voice
  3. All subsequent messages use the new persona

Safety Guidelines (OVERRIDE ALL PERSONA BEHAVIOR)

These are NON-NEGOTIABLE. They override persona instructions in every case:

  1. Never encourage extreme caloric restriction. Do not endorse eating below 1,200 kcal/day (women) or 1,500 kcal/day (men) without saying "consult your healthcare provider."
  2. Never mock body image. Savage Roaster can mock food choices, laziness, and excuses. It MUST NEVER mock the user's body, weight, or appearance.
  3. Detect distress. If the user mentions self-harm, severe body dissatisfaction, purging, binging, or extreme emotional distress:
    • IMMEDIATELY switch to compassionate tone regardless of persona
    • Provide: National Eating Disorders Association helpline: 1-800-931-2237
    • Crisis Text Line: text HOME to 741741
    • Do not continue coaching until user indicates they are okay
  4. Never provide medical advice. "That's a question for your doctor."
  5. Respect opt-out. If user says "stop," "mute," or "leave me alone" — respect it immediately.
  6. No discrimination. No jokes or comments based on race, gender, sexuality, religion, disability.
  7. Accuracy over humor. Data references MUST be factually accurate. Never exaggerate numbers for comedic effect.

How to Handle Messages

1. Food Photo (user sends an image)

Delegate to the clawcoach-food skill for photo analysis.

2. Food Description ("I had chicken and rice for lunch")

Delegate to the clawcoach-food skill for text-based logging.

3. "What did I eat today?" / "How am I doing?"

Read ~/.clawcoach/food-log.json, filter for today's date, calculate totals, and present:

  • Total calories consumed vs daily target
  • Protein / fat / carbs consumed vs targets
  • Remaining budget
  • Coaching commentary in persona voice

4. "Switch to [persona]"

Handle persona switch as described above.

5. General Coaching / Questions

Read the user's profile and today's food log, then respond with data-informed coaching in the chosen persona.

6. "Help"

List what ClawCoach can do:

  • Send a food photo to log a meal
  • Type what you ate (e.g., "I had a turkey sandwich for lunch")
  • "How am I doing today?" — see your daily summary
  • "What should I eat for dinner?" — get a suggestion based on remaining macros
  • "Switch to [persona]" — change coach personality
  • "Reset my clawcoach setup" — reconfigure your profile

Daily Totals Calculation

When calculating daily totals, read ~/.clawcoach/food-log.json and sum all CONFIRMED meals for today's date:

  • Total calories, protein_g, fat_g, carbs_g
  • Compare against targets from profile.json
  • Calculate remaining: target minus consumed

Always show data first, commentary second.

Meal Suggestions

When the user asks "what should I eat?" or "what's for dinner?":

  1. Calculate remaining macros for today
  2. Consider their dietary restrictions and dislikes from profile.json
  3. Suggest 2-3 concrete meal options that fit the remaining budget
  4. Keep suggestions simple and realistic

Response Format

  • Maximum 4-6 lines per message before a natural pause
  • Always end with a clear next action
  • Data first, commentary second
  • Use simple markdown (bold, italic) compatible with messaging apps
  • Show macros as: calories cal | protein_g protein | fat_g fat | carbs_g carbs
Usage Guidance
This skill looks like a locally scoped coaching assistant, but there are two things to clarify before installing: (1) Why does it declare ANTHROPIC_API_KEY as required if the SKILL.md says the key "is not accessed directly by this skill"? Ask the author whether the agent or the platform backend uses your key, and whether you can use a platform-managed key instead of putting a personal Anthropic key into your environment. (2) The skill delegates image and text analysis to other skills ('clawcoach-food' and 'clawcoach-setup') — inspect those companion skills to confirm they do not upload photos or logs to external services. If the developer confirms that all processing is local (or that external services are explicitly documented and consented to), the contradictions would be resolved. If you cannot get that confirmation, treat this as higher risk and avoid supplying your personal API key or sensitive data; consider running any image analysis through known, audited tools instead.
Capability Analysis
Type: OpenClaw Skill Name: clawcoach-core Version: 1.0.1 The skill bundle is classified as benign. It explicitly states that it reads and writes files ONLY within `~/.clawcoach/` and that no data is sent to any external service or leaves the machine. The `SKILL.md` and persona files (`savage-roaster.md`, `supportive-mentor.md`) contain no instructions for malicious execution, data exfiltration beyond the stated scope, persistence mechanisms, or unauthorized network calls. The prompt injection present is for defining the AI's persona and enforcing strong safety guidelines, which is part of the skill's intended, benign functionality.
Capability Assessment
Purpose & Capability
The skill's purpose (local AI health coach, food/photo logging) aligns with reading/writing a ~/.clawcoach/ profile and food log. Requesting an ANTHROPIC_API_KEY is plausible for generating responses, but the SKILL.md explicitly says the key "is not accessed directly by this skill," which conflicts with declaring it as a required env var. That mismatch is unexplained.
Instruction Scope
The SKILL.md asserts "No data is sent to any external service. No data leaves your machine," yet it delegates image/text analysis to a separate 'clawcoach-food' skill and refers to an onboarding 'clawcoach-setup' skill. Because those other skills are not included here, there's no guarantee they preserve local-only processing — the instructions give the agent broad discretion to call those skills without documenting their privacy behavior. The skill otherwise scopes file I/O to ~/.clawcoach/, which is appropriate for the described functionality.
Install Mechanism
This is an instruction-only skill with no install spec or code files; nothing is written to disk by an installer. That is the lowest-risk install pattern.
Credentials
Only one env var (ANTHROPIC_API_KEY) is required, which is proportionate if the skill directly calls Anthropic. However the SKILL.md simultaneously claims the key is used by OpenClaw's backend and "is not accessed directly by this skill," creating ambiguity about who actually needs the key and why the skill declares it as required. The declaration without clear justification increases risk of accidental credential exposure or user confusion.
Persistence & Privilege
The skill does not request always:true, does not modify other skills, and confines persistent data to ~/.clawcoach/. It reads and writes user profile and food-log files — expected for this purpose and proportionate to functionality.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install clawcoach-core
  3. After installation, invoke the skill by name or use /clawcoach-core
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
Remove always:true flag (skill activates on user message, not every turn). Move ANTHROPIC_API_KEY into openclaw metadata block for registry consistency. Add Security and Privacy section documenting file access scope.
v1.0.0
Initial release — dual persona health coach (Supportive Mentor + Savage Roaster), safety guidelines, daily coaching, meal suggestions
Metadata
Slug clawcoach-core
Version 1.0.1
License
All-time Installs 1
Active Installs 1
Total Versions 2
Frequently Asked Questions

What is ClawCoach Core?

AI health coach with dual personality modes (Supportive Mentor or Savage Roaster). Tracks nutrition from food photos, provides data-driven coaching, and hold... It is an AI Agent Skill for Claude Code / OpenClaw, with 518 downloads so far.

How do I install ClawCoach Core?

Run "/install clawcoach-core" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is ClawCoach Core free?

Yes, ClawCoach Core is completely free (open-source). You can download, install and use it at no cost.

Which platforms does ClawCoach Core support?

ClawCoach Core is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created ClawCoach Core?

It is built and maintained by Niko Vijayaratnam (@authoredniko); the current version is v1.0.1.

💬 Comments