Content
72%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A strong, dense reference-style skill that concentrates on genuinely non-obvious Snowflake pitfalls and gives mostly executable examples. Its weaknesses are the absence of validation/verification steps for risky pipeline and DDL operations, and a monolithic single-file structure that could offload detailed reference material to bundle files.
Suggestions
Add explicit validation checkpoints to pipeline workflows (e.g. query INFORMATION_SCHEMA.TASK_HISTORY or SHOW TASKS after ALTER TASK ... RESUME to confirm the task is running, and check SYSTEM$STREAM_HAS_DATA before relying on a stream).
Replace the 'SELECT ...' placeholder in the task example with a concrete executable INSERT ... SELECT so the snippet is copy-paste ready.
Move the Cortex AI function table and the Common Error Patterns table into references/ files (e.g. references/cortex-functions.md, references/error-patterns.md) and link them from SKILL.md to reduce the main-file footprint.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and rule-dense with no concept padding: every section delivers Snowflake-specific gotchas (colon prefix rule, '$spec$' delimiter, TO_FILE argument split, deprecated function list) and assumes Claude's competence. Every token earns its place. | 5 / 5 |
Actionability | Mostly executable SQL/Python snippets with concrete BAD/GOOD contrasts (e.g. the colon-prefix procedure pair and the MERGE example), but minor gaps remain: the task example uses a literal 'SELECT ...' placeholder and the stored procedure snippet omits the '$$' body delimiters. | 4 / 5 |
Workflow Clarity | The 'Choosing Your Approach' table and the 'Common Error Patterns' table provide a decision sequence and error-to-fix mapping, but there are no explicit validation checkpoints (e.g. verifying task state after ALTER TASK ... RESUME). Since the skill covers destructive/batch operations (CREATE OR REPLACE, pipeline refresh), the missing validation steps cap this at 3. | 3 / 5 |
Progressive Disclosure | Sections are clearly organized with headers and tables, but this is a single ~225-line file with no bundle references at all; content such as the Cortex function reference table and the error-pattern table could live in separate reference files. It is more organized than the anchor-2 'wall of text', but not the well-split anchor-4/5 structure. | 3 / 5 |
Total | 15 / 20 Passed |