[patil-prajwal]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

2. Or, if no issue exists, describe the change:

Problem:
_load_skill_from_dir eagerly reads the full markdown body of SKILL.md (L2 instructions) and recursively reads the content of all files in references/, assets/, and scripts/ (L3 resources) into memory at startup. For agents loading dozens of skills or large references, this introduces unnecessary I/O and memory overhead when only L1 metadata (name, description) is needed for initial registration.

Solution:

1. Parse only the YAML frontmatter of SKILL.md in chunks (_parse_skill_md_frontmatter) during discovery.
2. Scan resource directory entries (_scan_dir) without reading file contents into memory.
3. Defer reading L2 instructions and L3 resources until accessed, caching results in Skill / Resources on first access.
4. Maintain 100% backward compatibility for both attribute access (skill.instructions, skill.resources.references) and lookup methods (get_reference, list_references, get_script, etc.).

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Summary:

• Added test__load_skill_from_dir_defers_instructions_and_resources_until_accessed in tests/unittests/skills/test__utils.py verifying zero file reads occur at load time and reads only happen on access.
• Passed scripts/compliance_checks.py, ruff check, and all skills unittests.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.