Lists multiple specific concrete actions: prepare Azure apps for deployment, infrastructure via Bicep/Terraform, azure.yaml, Dockerfiles, create/modernize apps, add authentication, add caching, and more. Very detailed about what it does.

Clearly answers both 'what' (prepare Azure apps for deployment with infra Bicep/Terraform, azure.yaml, Dockerfiles) and 'when' with an extensive explicit WHEN clause listing numerous trigger phrases. Also includes explicit exclusions (DO NOT USE FOR) which further clarifies when to use it.

Excellent coverage of natural user phrases including 'create app', 'build web app', 'deploy to Azure', 'generate Terraform', 'generate Bicep', 'function app', 'todo list with frontend and API', 'static portfolio website', and many more variations users would naturally say.

Explicitly distinguishes itself from related skills by name: 'not cross-cloud migration (use azure-cloud-migrate)' and 'DO NOT USE FOR: copilot-sdk apps (use azure-hosted-copilot-sdk)'. This clear boundary-setting with named alternatives makes it highly distinctive and reduces conflict risk.

Extremely verbose and repetitive. The same instructions are stated multiple times — the plan-first requirement is repeated at least 4-5 times with nearly identical wording (Rules #1, the PLAN-FIRST WORKFLOW box, Phase 1 intro, Phase 1 Step 6). The mandatory hand-off to azure-validate is repeated 3+ times. The file name `.azure/deployment-plan.md` is mentioned over a dozen times. Much of this redundancy wastes tokens without adding clarity.

The skill provides structured steps with references to external files for concrete details, and includes specific file paths and tool names (azd, Bicep, Terraform). However, the skill itself contains no executable code examples, no concrete command snippets beyond mentioning `azd init -t`, and delegates nearly all concrete guidance to reference files that are not provided. The routing tables and phase tables are helpful but still abstract.

The two-phase workflow (Planning → Execution) is clearly sequenced with numbered steps, explicit blocking gates ('STOP HERE — Do NOT proceed to Phase 2 until the user approves'), mandatory validation checkpoints (update plan status before hand-off, invoke azure-validate before azure-deploy), and error prevention rules. Feedback loops are present (destructive actions require ask_user, validate before deploy).

The skill references many external files (references/analyze.md, references/requirements.md, etc.) which is good progressive disclosure structure. However, no bundle files were provided so we cannot verify these exist, and the main SKILL.md itself is very long (~200+ lines) with significant inline repetition that could be condensed. The SDK Quick References section is a good example of well-structured disclosure, but the repeated mandatory warnings bloat the main file.