SOE Compliant Office Agent Skill System: 20 Skills Across 3 Domains

SOE Compliant Office Agent Skill System: Full-Stack Practice of 20 Skills Across 3 Domains

Introduction: Why State-Owned Enterprises Need Dedicated Compliance Office Skills

There are plenty of AI office tools on the market -- writing documents, making PPTs, creating reports, the basic four-suite can handle it all. But in the context of State-Owned Enterprises (SOEs), three fatal problems immediately emerge:

1. Compliance Gap: Party committee pre-review, "Three Majors and One Large" decisions, GB9704 document format -- generic tools have no understanding of these regulations
2. Audit Blind Spots: Who used AI to change what and when? No traceability means inspection teams get blank stares
3. Standards Disconnect: Xinchuang (domestic IT) requirements, domestic model adaptation, penetrative supervision -- generic solutions avoid all of these

We spent 6 months building 20 Skills across a three-domain architecture -- not a fine-tuned version of generic office tools, but purpose-built from zero for SOE needs.

Three-Domain Architecture: Document Operations + Compliance Security + Reporting Analysis

Domain	Skills	Phase
Document Operations	doc-formatter, doc-template, doc-compare, meeting-minutes, contract-check, red-letter	Phase 1 (8 skills, done)
Compliance Security	party-review, triple-major, gb9704-check, audit-trail, data-mask, poc-filter	Phase 2 (6 skills, done)
Reporting Analysis	ops-digest, risk-weekly, kpi-dashboard, var-analysis, peer-compare, forecast	Phase 3 (6 skills, done)

Three Key Differentiators

1. Built-in Compliance

Not "check compliance after writing" but "write within a compliance framework from the start." The Red-Header Document Generator Skill, for example, validates every step against national standards -- layout position, copy numbering, classification markings, urgency levels -- all with corresponding GB/T 9704 validation rules.

2. Audit Trail by Default

Every AI operation's complete chain is recorded locally in the .audit/ directory:

• Who called which Skill and when
• What raw data was input (archived after desensitization)
• What intermediate results AI produced
• What final content was output
• Whether human confirmation was obtained

3. National Standards Ready

20 Skills incorporate 15 SOE-specific standard systems:

Standard	Skills	Implementation
GB/T 9704-2012	doc-formatter, gb9704-check, red-letter	147-item validation
SASAC EVA Assessment	ops-digest, kpi-dashboard	Automatic calculation
"Three Majors" Decision System	triple-major, party-review	Amount threshold trigger
SOE Procurement Compliance	contract-check	50 built-in rules
Penetrative Supervision	risk-weekly	Five-dimension framework
Xinchuang Requirements	All skills	Pure Python + domestic models

Technical Implementation: Zero API Cost + Pure Python + Millisecond Response

Key technical features:

• Response Speed: All rules run in-memory, single judgment < 10ms
• Zero API Cost: No LLM API calls for decisions, pure rule engine
• Auditable: Every judgment records rule ID, threshold, and actual value
• Configurable: Amount thresholds and rule switches via YAML config

`python

Core logic of the "Three Majors" determination Skill

def check_triple_major(item):
decisions = []
# Rule 1: Major decision -- investment exceeds 5% of net assets
if item.type == "investment" and item.amount > company.net_assets * 0.05:
decisions.append({
"rule": "Major Decision - Large Investment",
"threshold": f"5% of net assets ({company.net_assets * 0.05:.0f}万元)",
"actual": f"{item.amount:.0f}万元",
"action": "Must submit to Party Committee for pre-review"
})
return decisions
`

SOE Maturity Scoring: 1-5 Compliance Levels

Level	Description	Characteristics
SOE-1	Manual Compliance	Pure manual operation, compliance depends on personnel quality
SOE-2	Tool-Assisted	Generic office tools, compliance check via manual review
SOE-3	Built-in Rules	Compliance Skills with automatic validation, human confirmation
SOE-4	Audit Closed-Loop	AI full-chain traceability, compliance proactive alerts
SOE-5	Intelligent Compliance	AI auto-adapts to regulatory changes, compliance as default state

Most central SOEs currently sit between SOE-1 and SOE-2. This Skill system enables a direct leap to SOE-3 or even SOE-4.

Open Source & Cross-Promotion

Repository	Content	License	Link
soe-compliant-office	20 SOE compliance office Skills	MIT	https://github.com/yuzhaopeng-up/soe-compliant-office
financial-ai-skills	104 financial AI skills	MIT	https://github.com/yuzhaopeng-up/financial-ai-skills
teleagent-skills	5 general business skills	Apache 2.0	https://github.com/yuzhaopeng-up/teleagent-skills
skill-framework	208 skill taxonomy + L0-L4 framework	MIT	https://github.com/yuzhaopeng-up/skill-framework
fintech-h5-demos	12 zero-dependency financial H5 demos	MIT	https://github.com/yuzhaopeng-up/fintech-h5-demos

Conclusion: From "Can Use AI" to "Use AI Compliantly"

SOE digitalization isn't a capability problem -- it's a compliance problem. It's not "can we do it" but "can it withstand inspection."

These 20 Skills aim to make compliance the default option -- every operation within a rule framework, every audit trace traceable, every output meeting national standards.

The leap from SOE-1 to SOE-4 hinges not on technology, but on making compliance a built-in process rather than a post-hoc check.