From devops-tools
This skill should be used when the user asks to "analyze a skill", "break down a skill", "find what can be scripted", "separate deterministic from LLM parts", "extract scripts from a skill", "simplify a SKILL.md", or wants to identify which parts of a skill require LLM judgment versus mechanical automation.
How this skill is triggered — by the user, by Claude, or both
Slash command
/devops-tools:skill-analyzerThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Analyze any skill to classify each step as deterministic (scriptable) or requiring LLM judgment, then propose a script interface and a simplified SKILL.md.
Analyze any skill to classify each step as deterministic (scriptable) or requiring LLM judgment, then propose a script interface and a simplified SKILL.md.
Activate this skill when the user:
Default output: Provide the full analysis directly in the response, structured into the three sections described below (Classification Table, Script Interface, Simplified SKILL.md).
File input: When the user points to a specific SKILL.md file, read it and produce the analysis. When the user asks to apply changes, write the proposed script and simplified SKILL.md to disk.
Read the skill completely. Extract every discrete action, instruction, or decision point as a numbered step. Include implicit steps (e.g., "read the file" before "edit the file") that the skill assumes.
For each step, assign exactly one category:
| Category | Label | Definition | Examples |
|---|---|---|---|
| Deterministic | SCRIPT | Predictable outcome. No judgment, interpretation, or context-sensitivity needed. Inputs and outputs are well-defined. | Running a CLI command, string replacement in a file, creating a branch with a known name, formatting output to a fixed template, copying a file, reading a path from config |
| Requires LLM | LLM | Requires inference, interpretation, or context-dependent reasoning. The correct action depends on understanding meaning, not just following a rule. | Deciding which repo to target, judging if a user's request is complete, interpreting vague instructions, choosing between approaches, evaluating quality, summarizing content, generating prose |
Deciding between the two:
SCRIPT.LLM.LLM. It is better to keep judgment in the SKILL.md than to force a brittle script.Output a table with one row per step:
| # | Step | Category | Rationale |
|---|------|----------|-----------|
| 1 | Read the input file | SCRIPT | File path is provided, no interpretation needed |
| 2 | Identify the core message | LLM | Requires comprehension of meaning |
| ...| ... | ... | ... |
Group all SCRIPT steps into a proposed shell or Python script. The script should:
./scripts/prepare.sh <input>"Provide the full script source in a fenced code block. Include a brief comment header describing what it does and what inputs it expects.
Write a new SKILL.md that:
LLM steps as instructions for the LLM to followSCRIPT steps with a single instruction to invoke the script, e.g., "Run ./scripts/prepare.sh <path> to set up the workspace"End with a summary:
SCRIPT stepsLLM stepsA step is deterministic when:
Common deterministic operations:
A step requires LLM judgment when:
Common LLM-dependent operations:
For detailed examples of analyzing real skills:
npx claudepluginhub staticaland/skills --plugin devops-toolsCreates, fixes, analyzes, and validates skills for AI agents. Includes scripts for structure checks, quality analysis, and character budget management.
Creates, audits, and consolidates Claude Code agent skills following the Agent Skills open standard. Interviews for intent, scope, and edge cases before drafting.
Creates a SKILL.md file following the Agent Skills open standard (agentskills.io). Covers frontmatter schema, section structure, Expected/On failure pairs, validation checklists, and registry integration.