Ski Entity File Builder

• The name follows the kebab-case convention with the correct prefix for its type.
• The frontmatter description does not exceed 250 characters.
• Cross-reference paths use the correct relative format.
• The intricacy level determines the content density (see section 4).
• The expected size and density: if the projected content approaches or exceeds the recommended character limit for the entity type (<6000 Workflow/KB, <3000 Agent/Rule, <1500 Skill/Command), prepare to partition it by creating documents in the /resources directory and referencing them.

2. Naming conventions by type

3. Templates by entity type

Templates are split by entity category to minimize context load:

• Behavioral entities (wor-, age-spe-, age-sup-, com-): read ../../resources/res-entity-templates-behavioral.md
• Support entities (ski-, rul-, kno-, res-, scp-, hok-): read ../../resources/res-entity-templates-support.md

Extract the structure for the requested entity type and fill it dynamically per the assigned intricacy level.

4. Intricacy levels and cross-reference rules

Full specifications for all three intricacy levels (simple, medium, complex), cross-reference path conventions, consistency rules, content partitioning, and pre-validation checklist:

../../resources/res-entity-builder-protocol.md

Platform output: This skill generates GA (.agents/) entities only. Platform-specific output (CC, Codex) is handled by ski-output-claude-code and ski-output-codex respectively. Skills use the ski-[name]/SKILL.md subdirectory structure on all platforms — never create flat ski-name.md files.

5. Content Structuring and Partitioning (/resources)

If when planning the intricacy level (especially for complex) you anticipate a very extensive entity or one that will exceed the recommended limit:

1. Identify dense blocks that could be externalized (e.g. very long prompts, extensive categorization tables, few-shot examples, detailed style policies or guides).
2. Determine which support files to create in the ./resources/ directory to host that raw information.
3. In the main entity, make a direct reference to the support files structuring the information as a relational system. E.g. See detailed policies in [Security Policies](./resources/res-security-policies.md).

6. Pre-presentation validation

After generating the entity content and before returning it, run this automated checklist:

1. Frontmatter: name and description are present; description ≤ 250 chars.
2. Cross-references: Each path in Skills/KB/Rules tables points to an entity that exists or is planned in creation_order.
3. Character count: If the entity exceeds the recommended limit for its type, flag it with a suggestion to partition into resources/.
4. Template conformance: All required sections for the entity type are present and non-empty.
5. Naming: Correct prefix for the entity type and kebab-case format.

Emit a summary line at the end of the generated entity:

Pre-validation: ✅ frontmatter | ✅ cross-refs | ✅ size (2847/3000) | ✅ sections | ✅ naming

If any check fails, use ⚠️ and describe the issue. Do not suppress failures.

7. Universal compliance requirement for generated agents

Universal requirement (all intricacy levels): Every generated agent (type age-spe-* or age-sup-*) MUST include in its Execution Protocol:

"Before presenting your output, emit a <sys-eval> block per rul-strict-compliance."

Every agent's Related rules table MUST include rul-strict-compliance. This ensures all generated systems produce traceable outputs.

Examples

Example — Generation of Agent Specialist at simple level

Input:

{
  "type": "agent-specialist",
  "name": "age-spe-email-classifier",
  "function": "Classify incoming emails into predefined categories",
  "intricacy_level": "simple"
}

Expected output: Agent with Goals (2), Tasks (4), linear Execution Protocol (5-6 steps), Specific Rules (3), without Skills or KB if not needed.

Error Handling

• Name does not follow convention: Automatically correct and notify the user.
• Description exceeds 250 characters: Summarize while maintaining the essential meaning.
• Reference to entity not yet created: Include the reference with the correct path and indicate in a comment that entity will be created later.
• Inconsistency detected with the Blueprint: Pause, notify the user, and ask for clarification before continuing.