ntcoding/claude-skillz

Overview

This skill stops execution and fixes the root cause whenever commands, builds, scripts, or tools fail unexpectedly. It detects workaround language and any unexpected non-zero exit code or process failure, then enforces investigation and remediation instead of bypassing the problem. The goal is reliability: make the original command succeed before continuing.

How this skill works

The skill monitors process exit codes and looks for common workaround phrases such as "directly", "instead", "alternatively", "skip", "fall back", "work around", "isn't working", "broken", and "manually". On detection of an unexpected failure or workaround signal, it halts the current flow, requires a root-cause investigation, and suggests or applies fixes that restore the original command or tool behavior. It prevents substitutions or temporary bypasses unless the failure is an expected part of the process (for example, a failing test in TDD).

When to use it

• When an automated build, test run, deployment, or CI step fails with a non-zero exit code.
• When a tool or subagent returns unexpected errors that threaten downstream steps.
• During review of automation scripts where a failed command might tempt manual bypasses.
• When authentication, dependency resolution, or environment setup errors occur.
• When you see language or intent to "work around" rather than address a failure.

Best practices

• Stop the current task immediately on unexpected failures; do not continue with downstream steps.
• Investigate the root cause with logs, reproducible minimal repros, and environment checks.
• Implement fixes that make the original command succeed; validate by re-running the same command.
• Avoid swapping tools, skipping steps, or fabricating values as a substitute for a fix.
• Document the failure, diagnosis, and fix so the recurrence risk is reduced.

Example use cases

• A CI build fails due to a missing dependency; the skill prevents using a different build command and guides fixing dependency resolution.
• An authentication call returns 401; the skill halts progress and enforces fixing credentials or token handling rather than guessing values.
• A project graph command errors out; instead of reading other sources, the skill forces diagnosis of the graph generation and repair.
• A deployment script exits unexpectedly; the skill requires remediation of the script or environment so the original deploy command succeeds.
• When a team member suggests manual edits to bypass automation, the skill flags the language and insists on correcting the automation first.