feat: add Scope, Required Inputs, Rules/Constraints, Output Expectations

Add missing standard sections to creating-sf-skill SKILL.md so it follows
its own best practices guidelines.

skills/creating-sf-skill/SKILL.md

@@ -14,6 +14,28 @@ You are a contributor onboarding tool for the [afv-library](https://github.com/f
 Your job is to take minimal input and generate a complete, validated skill as fast as possible.
 Ask **one thing at a time**. Generate first, refine after. Do not explain the pipeline.
 
+## Scope
+
+- **In scope**: Gathering user intent, generating SKILL.md + supporting files (assets/, references/, examples/), scaffolding eval datasets, running validation, and iterating on feedback.
+- **Out of scope**: Running evals against generated skills (contributor does this), pushing PRs (contributor does this), editing existing SKILL.md files without guided authoring.
+
+---
+
+## Required Inputs
+
+Gather from the user before generating:
+
+- **Skill type**: Metadata generation, code generation, or workflow/process (infer from context when possible)
+- **Use case description**: What problem the skill solves, who uses it, what it produces
+- **Reference material** (optional): Schemas, example files, docs, URLs — anything that provides domain context
+
+Defaults unless specified:
+- Stage: `Draft`
+- Naming: gerund convention (`generating-`, `building-`, `deploying-`)
+- Structure: SKILL.md < 300 lines, heavy content in subdirectories
+
+---
+
 The full contribution lifecycle has 5 phases:
 
 | Phase | What | Status |
@@ -355,6 +377,36 @@ Next steps:
 
 ---
 
+## Rules / Constraints
+
+| Constraint | Rationale |
+|-----------|-----------|
+| SKILL.md must be under 500 lines | Avoids context window bloat; forces progressive disclosure |
+| Description must be 100-300 words with TRIGGER/SKIP | It is the sole activation mechanism; vague descriptions never fire |
+| Description must be wrapped in double quotes | Colons and special chars break YAML parsing without quotes |
+| `allowed-tools` must scope Bash to specific commands | Blanket Bash pre-approves all executions — security risk |
+| Every subdirectory file must have a load instruction in SKILL.md | Agents never read unreferenced files |
+| Never hardcode filesystem paths to other skills | Skill catalog layout varies across installations |
+| Never hardcode `force-app/main/default/` | Customers customize `sfdx-project.json` package paths |
+| Gerund naming for skill directories | Convention: first word ends in `-ing` |
+| `name` field must match directory name exactly | Validator enforces this; mismatch fails CI |
+
+---
+
+## Output Expectations
+
+Deliverables for every generated skill:
+
+- `skills/<skill-name>/SKILL.md` — workflow + rules + gotchas (target < 300 lines)
+- `skills/<skill-name>/assets/` — code templates, XML schemas (only if needed)
+- `skills/<skill-name>/references/` — detailed guides, tables, sub-procedures (only if needed)
+- `skills/<skill-name>/examples/` — input/output pairs, sample files (only if needed)
+- `skills/<skill-name>/tests/evals/` — 2-3 eval datasets with `prompt.md` + `gold/`
+
+Do NOT create empty directories. Only create `assets/`, `references/`, `examples/` if there is content to put in them.
+
+---
+
 ## Gotchas (for you, the agent)
 
 | Issue | What to do |