Context modules

Typed, directly addressable Markdown context that supplements a SeedSpec without turning every concern into a core protocol field.

Optional context with an explicit role

A context module points directly to its Markdown entrypoint. Its requiredtype helps tools group the material. The Markdown remains the source of meaning; the type does not impose headings.

Missing optional hints do not make a module unreadable.

A consumer can always open the declared path and interpret the Markdown.

Supported types

TypeRole
intentAdditional product or domain intent.
skillInstructions and optional tools for performing work.
behaviorExpected agent conduct.
evaluationA method for assessing a scoped claim.
policyRules about what is required, permitted, or forbidden.
referenceSupporting facts, examples, source material, or documentation.
implementation-profileA prose description of one implementation approach.

Declaration

context_modules:
  - id: cloudflare-worker-profile
    type: implementation-profile
    description: Guidance for implementing the package on Cloudflare Workers.
    path: context-modules/cloudflare-worker/PROFILE.md
    applies_to:
      stages: [implementation, verification]
      capabilities: [daily-summary]

  - id: agent-skill
    type: skill
    description: Procedure and tools for building the daily summary.
    path: context-modules/build-summary/SKILL.md
    format:
      id: io.agentskills.skill
      version: "1"
      url: https://agentskills.io/specification

id, type, description, andpath are required. format andapplies_to are optional.

Format information is descriptive

A format can identify an existing convention and optionally point to packaged documentation or an HTTPS reference. Validation never fetches the URL. Tools may use the information for presentation or ask before retrieving external material.

SeedSpec does not require a universal context ontology or adapter registry. Unknown external conventions remain ordinary readable Markdown.

Progressive disclosure

applies_to can limit a module to lifecycle stages, capabilities, or success criteria. This lets a harness load relevant material when needed without changing its authority.