thkt/brush

/brush

Route a vague prompt to the best skill, or enrich it with relevant rules and constraints when no skill fits.

Input

$1: rough prompt (required). If empty, output usage example and exit.

Execution

| Step | Action | Detail | | --- | --- | --- | | 1 | Classify | Identify task type from $1: refactor / bug-fix / feature / review / investigation / other | | 2 | Scan skills | Glob ~/.claude/skills/*/SKILL.md; read descriptions of 3–5 candidates that match the task type | | 3 | Route or enrich | See decision below | | 4 | Output | See format below |

Decision

Route (skill found)

A skill is a clear match when its "Use when" description covers the task.

Output the routing recommendation. Include any gap — what the skill does NOT cover — if the original prompt had scope the skill won't handle.

Enrich (no skill match)

Rules from ~/.claude/rules/ are already loaded in context. Select the relevant ones and make them explicit in the rewritten prompt.

| Category | Inject when | | --- | --- | | PRINCIPLES.md (Occam, DRY, YAGNI) | Refactoring, architecture decisions | | CODE_THRESHOLDS.md | Any code quality ask | | CODE_STRUCTURE.md | Structural or organization questions | | SECURITY.md | Auth, input handling, API endpoints | | PRE_TASK_CHECK.md | Broad implementation requests |

Only inject rules that are directly relevant. Do not add length for thoroughness.

Output Format

Route case

→ /skill-name [suggested args]

Covers: [what the skill handles from the original prompt]
Gap: [what it does not cover, if any]

Enrich case

Original:
> [verbatim input]

Enriched:
> [rewritten prompt]

Added:
- [rule or principle]: [what was injected and why]