[sunwrobert]

Summary

Adds the agent workflow skills so they work on any machine without an employer's plugin repo. skills/ goes from 12 to 94.

+ 50 upstream skills, installed with `gh skill install --pin`, byte-identical to their pins
+  5 personal skills copied unchanged (equivalence-review, makefaster, prose-pass, restate, thermo-fix-loop)
+ 27 personal skills ported to be company-agnostic
~  1 skill updated (ship-with-proof gains the newer gates)

The 27 company-agnostic ports:

• Working style and loops: solar-mode, solar-loop, solar-loop-scout, solar-burn, solar-climb.
• PR workflow: ship-with-proof, green-stack, holistic-sweep, before-after-ss, deslop-micro, review-loop, pr-watchdog-devin.
• Devin variants: sol-loop-devin, council-devin.
• Visuals: viz, bulk-review-html, spotlight, animated-explainer.
• Telemetry: telemetry-hunt, datadog-dashboard, prod-bug-bash.
• SEO: seo-aeo, seo-content, google-search-console, bing-webmaster, ga4, gauge, gauge-cli.

What the ports changed:

• Reviewers and channels are inputs. The reviewer, review channel, target site, Datadog site, search property IDs and credentials all come from the repo, the user or the environment.
• review-loop replaces three review-bot skills. It requests review for one PR or a stack, then optionally drives it to approval. It never merges.
• viz falls back to a neutral token set (Tableau 10 palette, indigo chrome) when a project has no tokens of its own. Brand fonts, logos and data were removed, and the example data is synthetic.
• bulk-review-html and animated-explainer embed a Latin subset of Inter (OFL, license included).
• Dropped: skills that only made sense inside the original company.

Repo changes:

• tsconfig.json, vitest.config.ts and .oxlintrc.json skip skills/. Several skills bundle their own TypeScript and tests with their own runtimes (viz uses bun:test, statechart uses node:test).
• .prettierignore keeps the formatter off the 50 upstream skills, so they stay identical to their pins.
• skill-sources.json records the source of every skill.
• The README has a loop that links every skill into ~/.agents/skills and ~/.claude/skills, skipping names that already exist.

Testing

• Upstream skills: all 50 reinstall byte-for-byte with gh skill install <repo> <path> --pin <sha>.
• Leak review: an independent reviewer checked every ported file, including scripts, fixtures and font name tables, for company names, IDs, hosts, secrets, data and brand assets. I fixed every finding: real data in a chart example, a PR number, product examples, and brand fonts.
• Leak grep: a final grep for the company name, its review bot, internal service prefixes, Slack IDs and the old font name finds nothing outside upstream skills.
• Links: every link from a ported skill to another skill resolves.
• Repo checks: bun run check, bun run check-types and bun run test:unit pass.
• Formatter: oxfmt --check . reports only executor/executor.jsonc, which already fails on main.
• viz: bun test gives 44 pass and 0 fail. The validator accepts all six examples and still rejects its rejection fixture.
• bulk-review-html: build_test.py passes, and the example renders to PNG with the new font.
• Scripts: py_compile and bash -n pass for the remaining scripts.
• README loop: tested in a temporary HOME. It linked all 94 skills and left an existing directory untouched.