project-skill-scripts

/project:skill-scripts

Analyze plugin skills to identify opportunities where supporting scripts would improve performance (fewer tokens, faster execution, consistent results), then optionally create those scripts.

When to Use This Skill

Use this skill when...	Use alternative when...
Analyzing skill improvement opportunities	Need to create a single script for a skill
Bulk script creation across plugins	One-off script for one specific need
Measuring coverage of scripts across portfolio	Script generation is already done

Context

• Plugin root: !git rev-parse --show-toplevel
• Total plugins: !find . -maxdepth 2 -name 'plugin.json' -type f
• Skills with scripts: !find . -name 'scripts/*.sh' -type f

Parameters

Parse $ARGUMENTS for:

• --analyze: Scan all skills, report candidates (default)
• --create <plugin/skill>: Create script for specific skill only
• --all: Analyze and create scripts for all high-scoring candidates

Execution

Execute this skill script analysis and creation workflow:

Step 1: Run analysis script

Execute analyzer to get structured data on all skills:

1. Run analyzer: bash "${CLAUDE_PLUGIN_ROOT}/skills/project-discovery/scripts/analyze-skills.sh" $(git rev-parse --show-toplevel 2>/dev/null || echo '.')
2. Parse output to identify:
   • Current coverage (skills with scripts)
   • High-scoring candidates (score >= 8)
   • Script type recommendations

Step 2: Analyze candidates (--create or --all modes)

For each candidate skill:

1. Read SKILL.md to understand the workflow
2. Identify script opportunity patterns:
   • Multiple sequential git/gh commands → context-gather script
   • Multi-phase workflow → workflow script
   • Project type detection + conditional execution → multi-tool script
   • Repeated command with different args → utility script
3. Evaluate benefit: >= 4 tool calls, consistency, error handling, reuse frequency
4. Skip if: single simple commands, interactive/creative, already well-structured

Step 3: Create scripts

For approved candidates:

1. Use standard script template with structured output (KEY=value, section markers)
2. Follow design principles: structured output, error resilience, bounded output, portable
3. Place in <plugin>/skills/<skill-name>/scripts/<script-name>.sh
4. Make executable: chmod +x <path>
5. Update SKILL.md with "Recommended" section referencing the script
6. Update modified: date in frontmatter

Step 4: Report results

Present findings:

• Current coverage (X/Y skills have scripts)
• Scripts created (plugin, skill, script, type, commands replaced)
• Remaining candidates (plugin, skill, score, type, recommendation)
• Next steps (test, commit)

Step 5: Commit changes

If scripts created:

feat(<affected-plugins>): add supporting scripts to skills

Include in body:

• Which scripts were created
• What they replace (token/call savings)
• Which SKILL.md files were updated

Examples

Analyze Only

$ /project:skill-scripts --analyze

Skill Scripts Analysis

Current Coverage: 5/191 skills have supporting scripts

Top Candidates:
  git-plugin/gh-cli-agentic         score=14  type=context-gather
  kubernetes-plugin/kubectl-debugging score=12  type=multi-tool
  testing-plugin/playwright-testing   score=10  type=workflow

Create for Specific Skill

$ /project:skill-scripts --create testing-plugin/playwright-testing

Analyzing testing-plugin/playwright-testing...
Found: 6 bash blocks, 3 phases, 12 commands

Creating scripts/run-tests.sh...
- Consolidates: test discovery, execution, report parsing
- Replaces: 5 individual tool calls
- Output: structured test results with file:line references

Updated SKILL.md with "Recommended" section.

Error Handling

Situation	Action
Skill has no bash patterns	Skip, report "no script opportunity"
Script already exists	Report existing, ask to overwrite
SKILL.md is read-only	Report error, suggest manual update
Plugin not found	List available plugins