home / skills / fusengine / agents / sniper-check

sniper-check skill

safe

This skill validates code quality quickly by running an isolated sniper-check workflow and returning a final report with zero main-context pollution.

npx playbooks add skill fusengine/agents --skill sniper-check

Review the files below or copy the command above to add this skill to your agents.

Files (1)

SKILL.md

2.1 KB

---
name: sniper-check
description: Use when validating code quality after modifications. Runs sniper agent in isolated forked context for clean, fast validation.
context: fork
agent: fuse-ai-pilot:sniper
user-invocable: true
argument-hint: "[file-or-directory]"
---

**Target:** $ARGUMENTS

# Sniper Check

## Overview

Quick code quality validation using the sniper agent in an isolated forked context. Executes the full 6-phase workflow without polluting the main conversation context.

| Feature | Detail |
|---------|--------|
| **Context** | Forked (isolated sub-agent) |
| **Agent** | sniper (Sonnet) |
| **Phases** | 6-phase code-quality workflow |
| **Result** | Only final report returns to parent |

---

## When to Use

| Scenario | Use |
|----------|-----|
| After code modifications | `/sniper-check src/` |
| Validate specific file | `/sniper-check path/to/file.ts` |
| Full project check | `/sniper-check .` |
| After refactoring | `/sniper-check src/components/` |

---

## Workflow

Execute the mandatory 6-phase `code-quality` workflow:

1. **PHASE 1+2 (PARALLEL)**: Launch both in parallel:
   - `explore-codebase` (Haiku) → Architecture discovery
   - `research-expert` (Sonnet) → Documentation verification
2. **PHASE 3**: Grep all usages → Impact analysis
3. **PHASE 4**: Run linters → Detect errors
4. **PHASE 5**: Apply corrections → Minimal changes
5. **PHASE 6**: Re-run linters → Zero errors

**CRITICAL**: Phases 1+2 must run in PARALLEL (two Task calls in one message).

---

## Critical Rules

| Rule | Reason |
|------|--------|
| Never skip phases 1+2 | Documentation-backed fixes only |
| Always run phases in order | Dependencies between phases |
| Zero linter errors | Non-negotiable exit criteria |
| Minimal changes only | Smallest fix necessary |

---

## Report Format

Return a validation report with:

- Architecture summary (from explore-codebase)
- Documentation references (from research-expert)
- Impact analysis table (usages, risk level)
- Errors fixed (critical, high, medium, low)
- SOLID compliance status
- Final linter status (must be ZERO errors)

Overview

This skill runs a focused code-quality validation using the sniper agent in an isolated, forked context. It performs a mandatory 6-phase workflow to find, analyze, and fix issues while keeping the main conversation context clean. The final output is a concise validation report returned to the parent process.

How this skill works

The skill forks an isolated sub-agent that executes the six-phase code-quality workflow: parallel architecture discovery and documentation research, usage grep, linting, targeted fixes, and a final lint pass. Phases 1 and 2 run in parallel as two Task calls in one message to ensure documentation-backed fixes. Only the final validation report is sent back to the parent, preventing intermediate noise.

When to use it

After making code modifications to validate changes quickly
When validating a specific file or directory before committing
After refactoring to ensure no regressions or hidden usages remain
For a full-project sanity check before release or merge
When you need documentation-backed fixes and minimal code edits

Best practices

Always run the skill after edits to guarantee zero linter errors before merge
Point the skill at the smallest relevant path to reduce runtime and noise
Do not interrupt the forked workflow; phases must run in order
Review the impact analysis before accepting automated fixes
Enforce minimal change principle: prefer targeted fixes only

Example use cases

Validate a single file: /sniper-check path/to/file.ts to catch lint and usage issues
Sanity-check a refactor: /sniper-check src/components/ to detect missed usages
Full repository sweep: /sniper-check . before a release to ensure clean linters
Pre-merge gate: integrate sniper-check as a CI step for code-quality validation

FAQ

What happens if linters still report errors after the run?

The workflow requires a final zero-error lint status; if errors remain, the report highlights them and the run is considered failing until fixed.

Why run phases 1 and 2 in parallel?

Running explore-codebase and research-expert in parallel speeds validation and ensures fixes are supported by both architecture context and documentation.