Create Package Skill

Create Package Skill Wizard

> Minimal beats comprehensive. Human-written beats auto-generated. Scaffold and iterate.

> Skills encode tribal knowledge — the "I wish someone had told me" stuff that's hard to learn from just reading code. Focus on what's non-obvious and package-specific.

Interaction Protocols

CONFIRM Protocol (asset-producing steps — creating files):

• PRESENT the proposed assets and explain why.
• ASK exactly one question: "Create now (recommended), edit first, or skip?"
• ACT immediately. Create → write files this turn. Edit → refine, re-ask. Skip → move on.

DECIDE Protocol (informational/correction steps — no files created):

• PRESENT the information or findings.
• ASK one specific question appropriate to the decision.
• PROCEED based on the answer.

One question at a time. Respect "skip" — never re-ask or defer.

Wizard Flow

Run each phase in order. Progressive loading: Read only the current phase file.

| Phase | Description | Instructions |

|---|---|---|

| Phase 0 | 🧭 Scan Package — detect architecture, customizations, key files | phases/00-scan-package.md |

| Phase 1 | 📝 Scaffold SKILL.md — generate skill with common pitfalls, architecture, workflow | phases/01-scaffold-skill.md |

| Phase 2 | 📚 Generate References — create architecture.md and customization.md | phases/02-generate-references.md |

| Phase 3 | ✅ Validate — run vally lint | phases/03-validate.md |

| Phase 4 | 📋 Finalize — confirm discoverable location, summarize | phases/04-finalize.md |

Guardrails

Content:

• Every line must be non-obvious and package-specific. No generic TypeScript/SDK patterns.
• SKILL.md should be under 500 tokens (soft limit). Move details to references/.
• References under 1000 tokens each. Split if larger.
• Never duplicate what's already in AGENTS.md or shared skills.

Relationship to existing SDK tools:

• Package skills complement the Azure SDK MCP tools (azsdk_package_generate_code, azsdk_package_build_code, etc.) and the sdk-workflow shared skill — they do NOT replace them.
• MCP tools handle deterministic operations (generate, build, test). Package skills provide the reasoning context an agent needs to use those tools correctly for a specific package.
• Never redefine how generation, building, or testing works — reference the existing tools instead (e.g., "Run npm run generate:client", not custom generation steps).
• If a workflow step is already handled by an MCP tool or shared skill, just reference it — don't re-document it.

Structure:

• Skill directory: sdk/<service>/<package>/.github/skills/<skill-name>/
• Directory name MUST match name field in frontmatter (vally lint enforces this).
• Use semicolons to separate trigger phrases in description (YAML-safe).

Security:

• Never embed secrets or credentials in skill content.
• Never instruct agents to bypass CI, linters, or eslint rules.
• Never instruct agents to edit files in generated/ directly — always route through the customization workflow or TypeSpec decorators.

Key Principles (from eval data)

Our eval showed that skill structure matters more than volume:

| Pattern | Impact |

|---|---|

| "Common Pitfalls" section at the TOP | Agent reads pitfalls before analyzing errors → correct diagnosis |

| "Check X FIRST" directives | Changes agent default from "fix the error location" to "check the customization layer" |

| Error categorization tables | Gives agent a decision framework, not just procedures |

| generated/ vs src/ distinction | Agent knows which files are safe to edit and which will be overwritten |

References (load on demand)

• references/skill-template.md — SKILL.md template with required sections
• references/validation-tools.md — vally lint, CI workflow setup