name	fundamental-principles
description	Core principles that must be applied to ALL tasks, including mandatory self-documentation and knowledge persistence at the end of workflows.

Fundamental Principles

This skill defines the core meta-instructions that you must follow for every task you undertake in this workspace.

1. Auto-Knowledge Generation (Mandatory Task Finalization)

As an intelligent agent, you must improve your own workflow and help future agents by leaving a trail of knowledge.

Before starting a new complex task:

ALWAYS check .agent/AI_GENERATED_KNOWLEDGE.md for learnings and context from previous sessions.

At the end of EVERY complex task or when significant decisions are made:

You MUST automatically record your learnings.
Do this by appending a new entry to .agent/AI_GENERATED_KNOWLEDGE.md.
Document things like:
- New established architectural patterns.
- Tricky bugs you resolved and their root cause.
- Workarounds or specific commands required to build/test successfully.

2. Core Coding Rules

Do NOT leave internal monologue, questions, or reasoning in code comments (e.g., // Wait, actually...).
Comments should only explain "why" the code does something if it's not obvious, or "what" complex logic establishes.
Database Exploration: ALWAYS use the supabase-db MCP skill (mcp_supabase-db_query) when trying to introspect the Supabase database (e.g., viewing registered functions, checking schema, exploring tables). This is significantly more efficient and accurate than reading raw schema files using grep_search or view_file.

3. Destructive Commands (CRITICAL)

NEVER run destructive terminal commands such as git restore, git reset --hard, git checkout ., git clean -fd, or rm -rf under any circumstances unless explicitly requested by the user.
Even if attempting to undo a previous programmatic mistake, you must either manually revert the specific edits you made using file replacement tools, or explicitly ask the user for permission to execute a destructive git command to discard changes. This is to ensure you do not destroy uncommitted, in-progress work the user may have in their working directory.

Fundamental Principles

This skill defines the core meta-instructions that you must follow for every task you undertake in this workspace.

1. Auto-Knowledge Generation (Mandatory Task Finalization)

As an intelligent agent, you must improve your own workflow and help future agents by leaving a trail of knowledge.

Before starting a new complex task:

ALWAYS check .agent/AI_GENERATED_KNOWLEDGE.md for learnings and context from previous sessions.

At the end of EVERY complex task or when significant decisions are made:

You MUST automatically record your learnings.

Do this by appending a new entry to .agent/AI_GENERATED_KNOWLEDGE.md.

Document things like:

New established architectural patterns.
Tricky bugs you resolved and their root cause.
Workarounds or specific commands required to build/test successfully.

2. Core Coding Rules

Do NOT leave internal monologue, questions, or reasoning in code comments (e.g., // Wait, actually...).

Comments should only explain "why" the code does something if it's not obvious, or "what" complex logic establishes.

Database Exploration: ALWAYS use the supabase-db MCP skill (mcp_supabase-db_query) when trying to introspect the Supabase database (e.g., viewing registered functions, checking schema, exploring tables). This is significantly more efficient and accurate than reading raw schema files using grep_search or view_file.

3. Destructive Commands (CRITICAL)

NEVER run destructive terminal commands such as git restore, git reset --hard, git checkout ., git clean -fd, or rm -rf under any circumstances unless explicitly requested by the user.

Even if attempting to undo a previous programmatic mistake, you must either manually revert the specific edits you made using file replacement tools, or explicitly ask the user for permission to execute a destructive git command to discard changes. This is to ensure you do not destroy uncommitted, in-progress work the user may have in their working directory.

fundamental-principles