Thanks for improving home-omakase-skill. This repo ships a single agent skill;
keep changes small, evidence-based, and safe.
Any change to the skill must preserve these invariants:
- Food-safety rules stay intact. Every raw-fish path must keep the sashimi-grade + cold-chain + at-risk-population rules. Do not soften or remove them.
- Never fabricate sourcing facts. No invented vendors, prices, or "safe to eat raw" claims for a specific product. Region-specific channels are examples, not guarantees — phrase them as "confirm sashimi-grade with the vendor."
- The description is "when to use", not a workflow summary. Keep the
SKILL.mdfrontmatterdescriptionfocused on triggering conditions; never recite the steps there (it makes agents skip the body). - Bump the version + CHANGELOG for any user-visible change
(
skills/home-omakase/CHANGELOG.md, frontmattermetadata.version, README badge). Use semver: patch = wording/fix, minor = new content/section, major = restructure. - Keep
SKILL.md≤ 500 lines. Heavy material goes inreferences/.
- Safety rules preserved (point 1–2 above)
- Description is when-to-use only
- Version + CHANGELOG bumped
- Both
README.mdandREADME.zh-CN.mdupdated if behavior changed -
SKILL.mdstill ≤ 500 lines
feat: add kaiseki-only course template
fix: correct sushi-rice vinegar ratio
docs: clarify uni-as-offal ambiguity in fish guide
chore: bump to 0.2.0
The install unit is skills/home-omakase/. The .claude-plugin/,
.codex-plugin/, and .cursor-plugin/ manifests only carry metadata — skill
content is auto-discovered from skills/.