feat: Add support for parsing full GitHub URLs

[hoducha]

Problem

Running agr add with a full GitHub URL fails:

❯ agr add https://github.com/user/repo/tree/main/skills/my-skill
Failed: Invalid handle '...': too many path segments (expected user/name or user/repo/name)

Users often copy and paste URLs directly from the GitHub web interface. Additionally, some shared skills are located outside of the repository's root directory. However, the parse_handle function only accepts the shortened URL format in the form of user/repo/skill.

Solution

Detects GitHub URLs and normalizes them to the standard handle format before the existing parsing logic runs.

Supported URL patterns:

https://github.com/user/repo/tree/branch/path/to/skill → user/repo/skill
https://github.com/user/repo/tree/branch → user/repo
https://github.com/user/repo → user/repo
Also handles /blob/... URLs, trailing slashes, and http://

A fast-path check ("://" not in ref) skips urlparse entirely for normal handles.

[kasperjunge]

Review feedback: slash-in-branch-name edge case

After deeper analysis, the skill_path[-1] approach is actually correct for the real-world use case — users copy a URL pointing to a skill directory, so the last segment is always the skill name, regardless of slashes in the branch.

Suggestion: Add a comment in _try_parse_github_url documenting this known limitation:

if len(path_parts) >= 4 and path_parts[2] in ("tree", "blob"):
    # Note: we can't reliably separate the branch name from the skill
    # path when branches contain slashes (e.g. feature/foo). Since skill
    # discovery is name-based (recursive SKILL.md search), we only need
    # the last segment as the skill name — which skill_path[-1] gives us
    # correctly even with slashed branches.
    skill_path = path_parts[4:]

Also consider adding a test that documents this works:

def test_url_with_slashed_branch(self):
    """Slashed branch names still extract the correct skill name."""
    h = parse_handle("https://github.com/user/repo/tree/feature/login/skills/sample")
    assert h.name == "sample"
    assert h.repo == "repo"

Otherwise the implementation looks good — the "://" fast-path skip and the overall integration approach are clean. 👍

🤖 Generated with Claude Code