fix(skills-index): give the offline builder a walk budget the full ClawHub catalog fits in

[AIalliAI]

What does this PR do?

Unblocks the skills index / docs-site deploys, which have failed on every run since 2026-06-10 06:22 UTC.

Root cause: the interactive 12-second CATALOG_WALK_BUDGET_SECONDS added in 105625d ("honour overall_timeout and bound ClawHub catalog walk", the #38459 hang fix) also applies to the offline index builder. The builder's walk-to-exhaustion over the ~50k-skill ClawHub catalog needs ~250 sequential pages (~260s at the observed ~1 page/s), so the 12s budget truncates it at ~16 pages. The EXPECTED_FLOORS health check from #42347 then correctly refuses to ship the degenerate index (clawhub: 1797 < expected floor 20000 … 3195 < 20000), failing skills-index.yml and deploy-site.yml ever since — the live docs site is frozen at the June 9 state and the watchdog keeps bumping #38240.

Evidence

• Last green index run: 2026-06-09 18:52 UTC; first red: 2026-06-10 07:41 UTC — the budget commit landed on main 2026-06-10 06:22 UTC, in between.
• First failing run logs clawhub: 1797 skills (12.9s) — the crawl stops at exactly the 12s budget; later runs report 3195, i.e. the count varies with network speed, which is a truncation signature, not a catalog shrink.
• Live repro: a 30s budget walks 5,788 skills; 12s walks ~3.2k — linear in the budget.

Fix: ClawHubSource now accepts an explicit per-instance catalog_walk_budget_seconds; scripts/build_skills_index.py passes 600s (≈2.3× the measured full-walk time). Interactive callers (browse/search cold start) keep the tight 12s default, so the #38459 hang fix is untouched.

Related Issue

Fixes #38240

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• tools/skills_hub.py — ClawHubSource.__init__(catalog_walk_budget_seconds=None) instance override of the class default
• scripts/build_skills_index.py — builder constructs ClawHubSource(catalog_walk_budget_seconds=600)
• tests/tools/test_skills_hub_clawhub.py — regression test: instance budget wins over an expired class-level deadline, walk completes and caches
• tests/scripts/test_build_skills_index_health.py — fake-source stub accepts the new kwarg

How to Test

1. pytest tests/tools/test_skills_hub_clawhub.py tests/scripts/test_build_skills_index_health.py -q → 18 passed
2. pytest tests/tools -k skills_hub -q → 166 passed, 1 skipped
3. Live: ClawHubSource(catalog_walk_budget_seconds=30)._load_catalog_index() gathers ~5.8k skills vs ~3.2k at the 12s default; with 600s the walk reaches cursor exhaustion (~50k) and the health check floor passes
4. After merge, gh workflow run skills-index.yml should go green and the watchdog probe on [skills-index-watchdog] Skills index is stale or degraded (degraded) #38240 should recover

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run the relevant tests and all pass (see How to Test)
• I've added tests for my changes
• I've tested on my platform: macOS

Documentation & Housekeeping

• I've updated relevant documentation — N/A (internal CI/index builder behavior)
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact — N/A (no platform-specific code)
• I've updated tool descriptions/schemas if I changed tool behavior — N/A

Screenshots / Logs

First failing run (skills-index #27261057718):

  clawhub: 1797 skills (12.9s)
ERROR: skills index health check failed — refusing to ship a degenerate index.
  clawhub: 1797 < expected floor 20000

Latest deploy-site failure (#27366487914): clawhub: 3195 < expected floor 20000

[alt-glitch]

Duplicate of #44500 (merged) — that PR already unblocked the skills index by letting the offline builder walk the full ClawHub catalog (unbounded max_items=0), fixing #38240. This change targets the same root cause via a catalog_walk_budget_seconds=600 override; the fix is already on main.