Coding Plus

Upgrade your AI agent with a senior-level engineering SOP focused on inspection, minimal diffs, and hard verification.

Install
cmdop skills install agensi-coding-plus

Coding Plus is a skill that installs a Model-Agnostic Standard Operating Procedure (SOP) into an AI agent, replacing ad-hoc code generation with a structured senior engineering lifecycle. The workflow has four stages: deep inspection of existing code before any changes, contract definition that locks down what a fix must achieve, minimal-diff implementation that preserves existing patterns rather than rewriting files, and hard verification with explicit success criteria.

A core feature is epistemic categorization. The agent is required to label every claim as Verified, Inferred, or Unknown, making the boundary between confirmed facts and assumptions explicit. This directly addresses what the skill’s authors call hallucination-led development, where a model edits files based on assumed rather than observed codebase state.

Each response includes a Residual Risk report that documents what was not tested, so a human reviewer knows exactly where to focus attention. The SOP is model-agnostic and language-agnostic: it is described as compatible with Node.js, Python, Go, and other ecosystems, and works whether the underlying model is a hosted API or a local instance.

This skill is appropriate when the goal is to add consistent engineering discipline to an agent that will be making code changes across a real codebase. It is not a tool-calling server and exposes no discrete tools; its value is entirely in the behavioral constraints it places on the agent’s reasoning and output format.

Use cases

  • Enforce inspect-first code review before an agent edits any source file
  • Generate minimal diffs that preserve existing codebase patterns instead of full rewrites
  • Produce structured residual risk reports after a coding task so reviewers know what was not tested
  • Categorize agent claims as Verified, Inferred, or Unknown to surface hidden assumptions
  • Apply consistent senior-level engineering output across different underlying models (GPT-4, Claude, local)
  • Define explicit engineering success criteria before implementation begins

When to use it

  • An agent will be making autonomous code changes to a production or shared codebase
  • The team wants traceability between what the agent verified and what it assumed
  • Reducing hallucination-driven regressions is a priority
  • The codebase spans multiple languages or frameworks and a single consistent SOP is needed

When not to use it

  • The task requires calling external APIs or databases — this skill provides no tool integrations
  • A lightweight, single-file scripting task where full lifecycle overhead is unnecessary
  • The agent platform does not support skill or prompt injection
  • Discrete MCP tool calls are required; this skill exposes no tools