Skilllibrary skill-description-optimizer

Rewrite a skill's description field to fix routing problems — undertriggering, overtriggering, or passive language that fails to match user intent. Use when a skill isn't being found by agents, evaluation shows low trigger recall, or batch-auditing descriptions for a library release. Do not use when the skill's scope is the problem (not just description text) or when the description is already performing correctly.

install

source · Clone the upstream repo

git clone https://github.com/merceralex397-collab/skilllibrary

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/merceralex397-collab/skilllibrary "$T" && mkdir -p ~/.claude/skills && cp -r "$T/01-package-scaffolding/skill-description-optimizer" ~/.claude/skills/merceralex397-collab-skilllibrary-skill-description-optimizer && rm -rf "$T"

manifest: 01-package-scaffolding/skill-description-optimizer/SKILL.md

source content

Skill Description Optimizer

The

description

field is routing logic — it determines when hosts invoke the skill. Bad descriptions cause undertriggering (skill never fires) or overtriggering (fires for wrong tasks).

Procedure

1. Read current description

Assess:

Length: Too short (< 20 words) → likely undertrigger. Too long (> 100 words) → likely diluted.
Style: Capability statement ("This skill can...") vs trigger statement ("Use when...")
Discriminating power: Could this description match 5 other skills? If yes, it's too broad.

2. Identify failure mode

Mode	Symptoms	Cause
Undertrigger	Skill never fires for intended tasks	Description too narrow, technical, or missing user vocabulary
Overtrigger	Fires for unrelated tasks	Description too broad, overlaps siblings, no DO NOT USE boundary
Passive	Inconsistent triggering	Describes what skill is, not when to invoke it

3. Extract discriminating signals

What words predict THIS skill and not siblings?

Specific task verbs (scaffold, audit, deprecate, package)
Domain terms (ticket, manifest, overlay)
Explicit names the user would say

4. Write optimized description

Structure (40-70 words):

Sentence 1: Action verb + most discriminating signal (lead with what the skill does)
Sentence 2: Context or problem it solves
Sentence 3: "Use when [specific trigger condition]"
Sentence 4: "Do not use [explicit boundary]"

Anti-patterns to avoid:

"This skill helps with..." (passive)
"A comprehensive tool for..." (marketing copy)
"Use this when creating, adapting, refining, installing, testing, or packaging..." (kitchen-sink — matches everything)

5. Check sibling conflicts

Read 2-3 similar skill descriptions. Does the new description discriminate correctly? Add a distinguishing clause if needed.

6. Validate with test prompts

Prompt	Type	Should Match?
"[specific task in scope]"	positive	YES
"[specific task in scope]"	positive	YES
"[task for a sibling skill]"	negative	NO
"[task for a sibling skill]"	negative	NO

Output contract

## Description Optimization: [skill-name]

### Failure Mode
[Undertrigger | Overtrigger | Passive]

### Before
"[original description]"

### After
"[optimized description]"

### Discrimination Check
| Similar Skill | Differentiated By |
|---------------|-------------------|
| [sibling] | [distinguishing signal] |

### Test Prompts
| Prompt | Expected | Matches? |
|--------|----------|----------|

Failure handling

Scope ambiguity at procedure level: Description can't fix this — recommend scope narrowing or skill splitting
No discriminating signals: Skill may be too generic — narrow scope first
Can't test routing: Create synthetic test prompts based on intended use