chore: execute catalog stocktake and drop legacy host name

[kuker24]

Runs the verdict protocol that PR #18 installed but never executed, and clears the currency findings it surfaced.

Stocktake

docs/stocktake-1.8.3.md records all 62 items, one verdict each: 57 KEEP, 2 COMPRESS, 3 UPDATE, 0 MERGE, 0 RETIRE. Tree measured at 46 model-invoked + 16 manual, matching vendor/skill-policy.json and vendor/skill-allowlist.txt.

MERGE is zero on evidence. Highest lexical description overlap is matt-code-review ~ interrogate at 0.359 Jaccard, under the 50% warn line already enforced in tests/test_skills.py. Every near pair is a fenced sibling.

RETIRE is zero because no no-skill baseline was run. A baseline cannot come from a session that already has the catalog loaded, so five candidates (research, prototype, grilling, wait-what, ask-matt) are recorded as blocked on the eval-harness A/B gate rather than retired on argument.

Currency

GrokBuild is the predecessor product name, surviving in five bodies while absent from README and docs. Removed where it named the host, including the ask-matt description that loads on every session start. Left intact where it names an image/video tool contract; renaming that has a wider blast radius and is filed as a follow-up.

Two contradictions against rules/00-routing.md also fixed: chrome-devtools-axi offered a $HOME/.grok/bin fallback against the "do not depend on ~/.grok" rule, and grill-with-docs offered /implement which the rule says does not exist.

Compression

• grill-with-docs 120 → 77. The whole body was duplicated behind an orphan overlay marker dating to the 1.0.0 bootstrap, referenced nowhere in the engine. The halves differed, so the result is a reconciled superset, not a truncation.
• codebase-design 115 → 97. Two ASCII diagrams restated adjacent prose.
• chrome-devtools-axi 75 → 66 as a side effect of deduplicating its contract block; counted once under UPDATE.

Left alone: to-tickets, adhd, emil-design-eng, full-performance-audit. Their length is numeric gates and sequencing judgment.

Also

README documents the optional stitch MCP that shipped in 1.8.2 but was never listed beside serena and exa. GitHub repo description updated to 62 routed skills; topics untouched.

Verification

• python3 -m unittest tests.test_skills — 10 OK
• python3 -m unittest discover -s tests — 315 OK (skipped=2)
• python3 tests/test_isolation.py — 2 OK
• python3 -m compileall -q lib tests — clean
• git diff --check — clean

The four FAILED CHECKSUM_MISMATCH-style lines in suite output are pre-existing negative-path log noise from test_release_artifacts; verified identical on clean main.

Not in scope

No new skills, no new MCP, no new dependencies. VERSION stays 1.8.2, no 1.8.3 heading opened, no tag, no release. Historical CHANGELOG numbers untouched (CHANGELOG.md diff is 6 additions, 0 deletions). The 28 skills that defer licensing to vendor/license-audit.json were not given frontmatter keys.

Follow-ups for a maintainer

1. Run the A/B gate on the five blocked candidates before any RETIRE.
2. Decide whether the GrokBuild image_gen tool-family name should be renamed catalog-wide or documented as intended.
3. impeccable frontmatter says Apache 2.0 while the audit file says Apache-2.0; the test normalizes, but aligning would drop a special case.