Contextus is a file-first Agent Skill for maintaining the current truth about a company, product, or project. It gives people and agents one readable source for product facts, marketing, brand, website, technical context, decisions, changes, and unresolved conflicts.
Install the skill:
npx skills add kostja94/contextus --skill contextusAsk the agent to create the first context:
Use contextus to create a product context for Acme Studio from the facts I provide.
The first meaningful result is a root contextus.md: a human-readable overview with machine-readable YAML frontmatter and links to any detailed context modules.
Existing material can be imported without changing its source files:
Use contextus to analyze these project documents, identify current facts,
conflicts, history, and gaps, then propose a Contextus project.
Do not modify the source documents.
A Context Project may describe an organization, one product, or a bounded delivery project. It can maintain complete project-specific context across:
- identity, positioning, users, scenarios, capabilities, and limits;
- pricing, commercial rules, markets, locales, and availability;
- marketing research, keywords, competitors, content, and channel decisions;
- brand terminology, voice, visual identity, and prohibited claims;
- website structure, pages, routes, interfaces, and user journeys;
- repositories, architecture, deployment, data ownership, and technical constraints;
- confirmed decisions, meaningful changes, open questions, and useful history.
Contextus does not copy live operational data already owned by code, configuration, or another maintained system. It records the durable rule, necessary interpretation, and source location.
Every adopted project has one canonical entry at its repository root:
contextus.md
Its frontmatter identifies the subject, scope, visibility, format version, and maintained module paths. Its body gives people and agents a concise current overview. Detailed modules own their topics; the entry does not duplicate their full content.
Legacy project-context.md files and tool-specific paths may still be read as fallback material. They are not new-project templates and do not become parallel context systems.
The contextus skill supports seven operations through one entry point:
| Operation | Result |
|---|---|
| Initialize | Create context from explicit team or user knowledge |
| Import | Classify existing material without overwriting its source |
| Update | Change current truth and record the semantic transition |
| Resolve | Compare conflicting claims and request the missing judgment |
| Maintain | Find stale, missing, duplicated, or broken context |
| Export | Generate the minimum context required for one task and audience |
| Snapshot | Preserve an important task context when reproduction matters |
Explicit team statements can update current context directly. Agent research, code observations, project documents, and public sources remain candidate facts until a person confirms them. Unknown information stays unknown instead of becoming plausible copy.
Contextus distinguishes lifecycle states such as current, planned, experimental, deprecated, historical, and disputed. It also filters public, internal, and confidential information for each output.
Product truth changes as teams launch and retire capabilities, alter pricing, rebrand, expand markets, and pivot. Contextus keeps a ready-to-use current state while recording attributable, comparable, and reversible semantic changes.
Team statements, agent research, documents, code observations, and public sources act as structured contribution channels. Contextus normalizes their claims, sources, dates, status, and visibility. When inputs disagree, it exposes a branch-like semantic difference and asks for resolution instead of silently selecting or blending a version.
The analogy stops at the useful mechanics. Contributors are not reduced to APIs, and ordinary users do not manage branches or rebases.
- Marketing Skills consumes project truth for SEO, content, channels, and strategy work.
- Pagina consumes project truth while building pages and returns confirmed facts or candidates.
- Bricks consumes relevant product, interface, brand, and technical context while building components.
These projects remain usable without Contextus when the user supplies sufficient project material. They do not create or maintain another durable context system. The exact downstream behavior is maintained in the integration contract.
Contextus manages durable project context. It does not:
- make product or business strategy on the team's behalf;
- treat an agent inference as a confirmed fact;
- modify websites, repositories, campaigns, or other downstream assets without a separate explicit task;
- store secrets, credentials, evidence archives, or undigested meeting transcripts;
- act as personal memory, an enterprise approval platform, or a general document warehouse.
A hosted service, full CLI, database, and visual interface are intentionally deferred while the file model and Agent Skill workflow are proven.
skills/contextus/
SKILL.md
agents/openai.yaml
references/
information-model.md
workflows.md
integrations.md
templates/
organization.md
product.md
project.md
schemas/contextus-manifest.schema.json
examples/
scripts/validate.mjs
The information model owns context boundaries. Workflows own lifecycle behavior. Templates are adaptable starting points rather than mandatory document sets.
Run the repository checks after changing the skill, templates, schema, references, or examples:
npm test