Fix ClawHub install rate limiting by preferring /file endpoint

[mccar2cm]

Summary

• ClawHub's /download endpoint has a 20 req/window rate limit shared across IPs, causing skill installs to fail with 429 even on the first attempt (common on shared hosting/VPS)
• The /file endpoint has a 120 req/window limit and returns SKILL.md content directly — sufficient for the majority of ClawHub skills
• Now tries /skills/{slug}/file?path=SKILL.md first, falls back to /download only for non-SKILL.md packages (zips, Node.js)
• Respects the retry-after header on 429 responses instead of fixed exponential backoff
• Dashboard now shows the actual server error message on 500s instead of the generic "Server error — check daemon logs"

Test plan

• Install a SKILL.md-based skill from ClawHub (e.g., agentmail) — should succeed via /file endpoint
• Install a non-SKILL.md skill (zip/package) — should fall back to /download
• Verify cargo test --workspace passes (1886 tests)
• Verify cargo clippy --workspace --all-targets -- -D warnings is clean
• Trigger a 500 error in dashboard and verify the real error message is shown

🤖 Generated with Claude Code

[Avey777]

#492

[jaberjaber23]

thanks for this! we already implemented ClawHub rate limit retry with 5-attempt exponential backoff + jitter in v0.3.44 (fix for #492). appreciate the contribution

[Avey777]

thanks for this! we already implemented ClawHub rate limit retry with 5-attempt exponential backoff + jitter in v0.3.44 (fix for #492). appreciate the contribution

Is it because the project is still in the early stages? Would it be better to accept community contributions?

[mccar2cm]

No worries - I liked the idea of hitting the files endpoint then the download. Take a look, see if you think that is an interesting approach.

[Avey777]