Content
50%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.
The body is rich in executable gemini command examples and takes safety seriously (validation, rollback, sandbox guidance), but it is a 560-line monolith with substantial repetition and generic bash re-teaching that inflates token cost. Batch/destructive workflows lack embedded validation steps, capping workflow clarity.
Suggestions
Consolidate the five sections that repeat near-identical 'gemini --yolo -p' invocations into one command-pattern reference, and move the generic bash wrappers (retry loops, trap-based recovery, audit/md5 logging) out of SKILL.md entirely — Claude already knows bash, so keep only the Gemini-specific flags and semantics.
Split the executable scripts (backup_and_execute, audit_tools, with_rollback, tool-policy.json) into a scripts/ or references/ file and leave short, clearly signaled pointers in SKILL.md, turning the monolith into a lean overview.
Add explicit validation/verification steps inside the batch workflows (Bulk Processing, File Organization) — e.g., run a check or diff after moving files and updating imports, and only proceed on success — rather than relying on the standalone Safety Patterns section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~560-line body is noticeably verbose: near-identical 'gemini --yolo -p "..."' invocations recur across five separate sections (Available Tools, Execution Modes, YOLO Automation Patterns, Tool Workflows, Advanced Patterns), and entire sections (Monitoring & Logging, Error Handling) re-teach generic bash — retry loops, 'trap ERR', md5sum diffs, git stash — that Claude already knows. Only a small fraction (the flags --yolo, --sandbox, --checkpointing, --include-directories) is knowledge Claude does not already have. This matches anchor 2 ('several unnecessary explanations or padded sections'), not anchor 3, because the padding is whole sections rather than occasional sentences. | 2 / 5 |
Actionability | Mostly executable, copy-paste-ready guidance: concrete gemini commands, complete bash wrapper functions, and a tool-policy.json config. Minor gaps exist — the JSON policy block contains '//' comments making it invalid JSON as written, and quoted tildes ('~/.gemini/tool-usage.log', '~/.gemini/audit/...') do not expand in bash so the logging/audit scripts would write to a literal '~' directory — fitting anchor 4 ('concrete code or commands with minor gaps') rather than anchor 5. | 4 / 5 |
Workflow Clarity | Sequences are clear (numbered steps in project setup, code-generation pipeline, tool chaining), and the Safety Patterns section includes pre-execution validation and rollback. However, the batch workflows (Bulk Processing, File Organization moving files and rewriting imports, parallel execution) contain no validation/verification steps, and per the judging guidelines this caps workflow_clarity at 3 even though the raw sequencing would otherwise merit a 4. | 3 / 5 |
Progressive Disclosure | Section headers are clear and well-ordered, but the entire skill is a single inline monolith with no bundle files (no references/, scripts/, or assets/ exist). Content that clearly belongs in separate files — the backup/audit/retry shell functions and the tool-policy config — is inlined in SKILL.md rather than split out with pointers, matching anchor 3 ('content that should be separate is inline') better than anchor 2 (which requires minimal structure) and below anchor 4 (which expects appropriate splitting into referenced files). | 3 / 5 |
Total | 12 / 20 Passed |