An Agent Skill for turning company strategy, founder beliefs, talent choices, and real business conflicts into an actionable company SOUL and a decision system that both people and AI agents can execute.
It produces more than mission/vision/value copy. The skill builds and tests:
- mission and vision;
- worldview and human assumptions;
- talent philosophy and observable evidence;
- values and counter-behaviors;
- decision, product, and business principles;
- real-case tests for commercial and decision principles;
- one aligned human/agent decision kernel with priorities, exceptions, authority, stop rules, and verification;
- case-based evals that detect when people and agents interpret the same principle differently;
- 30/60/90-day culture implementation tied to business pain and growth.
The core method is simple: when foundational mission, vision, and values already exist, stop polishing slogans. Use real cases to test whether commercial and decision principles select a clear action, then compile the principles that survive into one semantic source of truth with a memorable human view and an executable agent view.
With a skills-compatible installer:
npx skills add binbingoes/design-company-soulManual installation:
git clone https://github.com/binbingoes/design-company-soul.git
cp -R design-company-soul/design-company-soul ~/.codex/skills/Other runtimes can copy the design-company-soul/ folder into their supported skills directory.
Use $design-company-soul to turn our company context and three real business conflicts into a SOUL v0.1 with one aligned human/agent decision kernel.
Chinese triggers include: 公司SOUL、企业文化、使命愿景、人才观、价值观、决策原则、人机决策原则、Agent SOUL、文化落地.
Validate a generated Markdown draft:
python design-company-soul/scripts/validate_soul.py draft.md --prettyThe validator checks the minimum structure, evidence hygiene, counter-behaviors, accountability, and stop-rule language.
The repository contains only generalized methods and a fictional example. Do not publish internal cases, personal data, customer details, unreleased metrics, or meeting transcripts without authorization.
MIT