agent-factory

schema	fcvw/skill@1
name	agent-factory
description	Evidence-based gate for creating a new skill or agent profile.
version	1.1.1
trigger_keywords	["create skill","create agent","new skill","nova skill","criar agente"]
session_types	["ai_governance","planning"]

SKILL: Agent and Skill Factory

Purpose

Govern the creation of new skills and specialized agent profiles without allowing uncontrolled proliferation. This skill is a Markdown-only gate: it decides whether a recurring problem deserves a reusable procedural asset, a small checklist inside an existing document, or no new asset.

Activation Triggers

Load this skill before creating any new file under skills/ or any new agent profile when:

• a user asks to create a new agent, skill, specialist, persona, command pack, or operational procedure;
• the same task pattern appears repeatedly in plans, sessions, troubleshooting records, or changelogs;
• an existing skill is too broad, too vague, or missing a repeatable procedure for a high-risk task;
• a proposed solution would otherwise add long instructions to AGENTS.md, AI.md, or another base-loaded document;
• an agent proposes a new role to solve a specific recurring problem.

Inputs

Recurring task evidence, existing skill/catalog coverage, trigger and responsibility boundary, expected consumers, validation task, maintenance owner, and token/risk ROI estimate.

Decision Ladder

Use the smallest reusable asset that solves the problem:

Level	Use when	Output
Inline checklist	One-off or low-risk clarification.	Add 3-7 bullets to the existing canonical document.
Template	Repeated output format is needed, but execution logic is simple.	Add a governance/TEMPLATE_*.md file.
Skill	A repeatable procedure has clear triggers, gates, validation, and exit criteria.	Add skills/<name>/SKILL.md.
Agent profile	A bounded role must execute a specialized loop with tools, journals, or handoff.	Add or update a focused agent-* skill/profile.

Creation Gate

Creation is allowed only when all required metrics pass:

Metric	Pass threshold
Recurrence	At least 2 independent occurrences, or 1 P1/P2 high-risk blocker.
Coverage gap	Existing docs/skills cover less than 70% of the needed procedure.
Token ROI	Expected initial-context reduction of at least 20%, or avoids adding 500+ words to base-loaded docs.
Risk ROI	Reduces a recurrent failure class, compliance risk, data risk, monolith risk, or validation gap.
Scope narrowness	One trigger family, one responsibility, one primary output, and one validation path.
Validation path	The new asset can be tested against at least one real plan/session/task.

If any metric fails, do not create a new skill or agent. Patch an existing document or defer with evidence.

Agent vs Skill Criteria

Create a skill when the asset is a procedure any agent can execute.

Create an agent profile only when all extra criteria pass:

• the work needs a persistent role boundary, not only a checklist;
• the role has a narrow objective and explicit non-responsibilities;
• the role can produce a measurable artifact or decision;
• the role does not duplicate orchestrator, agent-aegis, agent-hermes, or agent-hephaestus;
• the role has a journal or handoff rule when it can run independently.

Do not create persona-only agents, motivational roles, vague "expert" agents, or agents whose only difference is tone.

Block Conditions

Stop creation when any condition is true:

• the problem is a one-off request with no P1/P2 risk;
• the proposed asset overlaps more than 50% with an existing skill;
• the trigger cannot be stated in concrete keywords or session types;
• the asset has no validation method;
• the asset would encourage broad "do everything" execution;
• the content is mostly style, branding, naming, or personality;
• the proposed instructions should instead be a template field or a plan checklist.

Output Required

For every allowed creation, create or fill governance/TEMPLATE_AGENT_OR_SKILL_PROPOSAL.md and record the compact block below in the active plan:

## Agent/Skill Creation Gate

- Skill loaded: `skills/agent-factory/SKILL.md`
- Proposed asset:
- Asset type: `inline checklist` / `template` / `skill` / `agent profile`
- Evidence of recurrence:
- Existing coverage checked:
- Token ROI:
- Risk ROI:
- Scope boundary:
- Validation task:
- Decision: `create` / `patch existing` / `defer`

Exit Criteria

Creation is complete only when the asset has triggers, non-responsibilities, validation criteria, catalog/index updates, changelog entry, and at least one recorded use case.