SemanticSkills

Huawei Community Edition · community-maintained, not official Huawei Cloud.

Cloud agents get billing and pricing wrong in repeatable ways: wrong grain, guessed quotes, skipped reconciliation, answers with no evidence—or tables that break in Feishu and WeChat. SemanticSkills (Huawei Community Edition) packages Agent Skills where domain semantics live in references/, not in a bloated SKILL.md, so the agent picks the right fact entity before it runs a CLI.

中文说明见 README-CN.md. Repo release: v3.0.1 · see CHANGELOG.md (skill histories under qa/<name>/CHANGELOG.md).

Design

Ontology-first, skill-specific models. Each skill defines facts, grain, dimensions, and read-only operations in references/semantic/*.yml. Command templates stay in related-commands.md. SKILL.md holds safety rules, the workflow, and how to deliver answers (briefing-style or quote lines—IM-safe text, not pipe tables in chat).

Two skills today (independent install bundles; do not cross-route):

Skill	Concern	Primary ontology	Main BSS ops
huawei-cloud-billing-scout	Past spend — balance, bills, reconciliation, coupons	billing-ontology.yml	53 read-only query ops
huawei-cloud-cost-estimation	Future quotes — period & on-demand RFQ	rfq-period-model.yml, rfq-ondemand-model.yml	ListRateOnPeriodDetail, ListOnDemandResourceRatings

Shared routing pattern:

User question
     │
     ▼
catalog.yml ──────► route by pricing_mode or entry_point / triggers
     │
     ▼
semantic/*.yml ────► fact + evidence_boundary + dimensions
     │
     ▼
related-commands.md ► minimal read-only hcloud CLI (no --help first)
     │
     ▼
SKILL.md delivery ──► conclusion first · labeled basis · IM-safe bullets

Layer	Role	Typical files
Routing	Entry points and triggers	references/semantic/catalog.yml
Ontology	Facts, scope, evidence limits	billing-ontology.yml or rfq-*-model.yml + rfq-shared-dimensions.yml
Commands	Parameter templates, traps	related-commands.md, cli-installation.md, iam-policies.md
Protocol	Workflow, output contract	SKILL.md

skills/<name>/ is the install payload (npx skills add copies only this tree). qa/<name>/ holds validate.sh, evals, and audit configs (bin/gate.py where present, skillcheck.toml, etc.)—never installed with the skill.

Authoring: docs/authoring.md. Interaction discipline (一次只问一事): authoring § Interaction discipline.

Repository layout

SemanticSkills/
├── skills/<name>/       # SKILL.md + references/ (install bundle)
├── qa/<name>/           # validate.sh, evals/, bin/gate.py (optional), lint configs
├── docs/                # catalog.yml, authoring, per-agent install notes
├── tools/               # validate-all.sh, skill-scaffold.sh, install-git-hooks.sh
├── .githooks/           # pre-commit → validate-all.sh (set via install-git-hooks.sh)
├── template/{skill,qa}/
├── *-workspace/         # Skill Creator eval output (gitignored)
├── .agents/             # Local npx skills add copies (gitignored)
└── .credentials/        # Local credential samples (gitignored)

Skills

Skill	Version	Summary	Docs
huawei-cloud-billing-scout	2.3.8	Huawei Cloud Read-Only Billing — Spend, Charges & Reconciliation — one-page BSS briefing via KooCLI	details · changelog
huawei-cloud-cost-estimation	1.0.0	Huawei Cloud Pre-Order Cost Estimation — period and on-demand quotes via hcloud BSS	details · changelog
semantic-creator	0.2.0	Semantic Creator — meta-skill: interface → governed Kimball star-schema semantic layer (optional OKF export)	details · changelog

Index: docs/catalog.yml.

Install

GitHub (Cursor, Claude Code, Codex):

# Billing (past spend / reconciliation)
npx skills add ontology-of-everything/SemanticSkills \
  --skill huawei-cloud-billing-scout \
  --agent cursor \
  --copy -y

# Pre-order pricing (period / on-demand quotes)
npx skills add ontology-of-everything/SemanticSkills \
  --skill huawei-cloud-cost-estimation \
  --agent cursor \
  --copy -y

Replace --agent cursor with claude-code or codex as needed. List skills:

npx skills add ontology-of-everything/SemanticSkills --list

Local path (development):

npx skills add ./skills/huawei-cloud-cost-estimation \
  --skill huawei-cloud-cost-estimation \
  --agent cursor \
  --copy -y

Hermes (Hermes agent): hermes skills install ontology-of-everything/SemanticSkills/<skill-name> -y, or rsync ./skills/<name>/ into ~/.hermes/skills/.

Validate

Install local pre-commit (runs ./tools/validate-all.sh on every commit):

./tools/install-git-hooks.sh

All skills (same as CI):

./tools/validate-all.sh

Per skill:

./qa/huawei-cloud-billing-scout/validate.sh      # full gate: layout, ops, protocol eval, style
./qa/huawei-cloud-cost-estimation/validate.sh    # layout, skills-ref, markdownlint, skillcheck

Style-only audit (billing-scout):

python3 qa/huawei-cloud-billing-scout/bin/gate.py style

Optional real BSS smoke test:

HUAWEICLOUD_BILLING_SCOUT_REAL=1 \
HUAWEICLOUD_BILLING_SCOUT_CYCLE=2025-04 \
./qa/huawei-cloud-billing-scout/validate.sh

Offline protocol benchmark (Skill Creator layout):

# viewer: huawei-cloud-billing-scout-workspace/iteration-1/benchmark-review.html

Contributing

docs/CONTRIBUTING.md · docs/authoring.md

New skill:

./tools/skill-scaffold.sh <skill-name>

Sync on every skill change: skills/, qa/ (VERSION, CHANGELOG.md), docs/catalog.yml, docs/skills/<name>.md.

Marketplaces

Channel	Notes
skills.sh	Listed; install via npx skills add ontology-of-everything/SemanticSkills --skill <name>
SkillsMP	Repo topics claude-skills, claude-code-skill; SKILL.md frontmatter; crawler sync is periodic
ClawHub	clawhub skill publish; metadata.openclaw; skill bundle is MIT-0 on ClawHub (repo source Apache-2.0)
Cursor / Claude Code / Codex	npx skills add ... --agent <agent>
Hermes	hermes skills install or copy skills/<name>/ — see docs/agents/hermes.md

Reference publish flow: docs/skills/huawei-cloud-billing-scout.md.