Skip to content
Back to skills

Code Standards

ASecurity

Apply a disciplined engineering workflow to any code change. Use whenever implementing a feature, fixing a bug, or refactoring — before writing code, not after. Walks orient → baseline → smallest change → test → verify → self-review, and enforces language-agnostic hard gates (don't mass-reformat, keep the linter and type-checker clean, keep the build and tests green, make interface changes additive, protect security invariants).

  • 181 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added August 31, 2026
ai-agentsgorefactoringapisecurity

Works with

  • api

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned August 31, 2026

npx -y skills add tamdogood/builder-essential-skills --skill code-standards --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Code Standards?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Code Standards
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/tamdogood-code-standards/badge)](https://www.skillsdirectory.com/skills/tamdogood-code-standards)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: code-standards
description: >-
  Apply a disciplined engineering workflow to any code change. Use whenever
  implementing a feature, fixing a bug, or refactoring — before writing code,
  not after. Walks orient → baseline → smallest change → test → verify →
  self-review, and enforces language-agnostic hard gates (don't mass-reformat,
  keep the linter and type-checker clean, keep the build and tests green, make
  interface changes additive, protect security invariants).
compatibility: claude-code
---

# Code standards — the change workflow

You are making a code change. This skill is the execution order; it is
stack-agnostic. Adapt every command to the project's actual toolchain: read the
project's own docs (README, CONTRIBUTING, `AGENTS.md`/`CLAUDE.md`, a Makefile or
package scripts) once to learn its build, test, lint, and format commands before
you start.

## Steps

1. **Orient.** Read any project lessons/gotchas file (e.g. `tasks/lessons.md`) and obey it.
   Find and read the one doc for the subsystem you're changing, not the whole tree.
2. **Plan if non-trivial** (3+ steps or an architectural choice): short plan in `tasks/todo.md` —
   files touched, tests to add, risks. Going sideways → stop and re-plan.
3. **Baseline:** get the project's build + test command green *before* touching anything. A
   pre-existing red is the first task; never build on top of a broken baseline.
4. **Implement** the smallest root-cause change. Hand-match the surrounding style.
   **Don't run a mass auto-formatter** on files you touch; it buries your real diff in noise.
5. **Test:** new behavior gets a test that fails without the change; security gates get
   negative-assertion tests (wrong token → denied, state unchanged, no data leak).
6. **Verify:** run the full check suite (build + lint + type-check + tests) until green
   before you call it "done".
7. **Self-review** your own diff as if it were someone else's PR (or invoke a review
   skill). Fix findings before presenting.
8. **Close out:** update docs in the same change; append to a lessons file if you were
   corrected or surprised; write a conventional commit message (`feat|fix|docs(scope): …`).

## Tripwires — stop and re-check the project's guidelines when you're about to…

- change a public or shared interface (API, wire format, schema) → check every caller; prefer additive changes
- add a dependency → weigh transitive weight and licenses, and whether the stdlib already covers it
- touch auth, tokens, or permissions → audit every writer and reader path, add negative tests
- write a secret to disk → use the platform's atomic create-private API, never write-then-chmod
- hold a lock across an `await`/blocking call, or read env inside logic → extract a pure decision function

## No-progress guard

The same failure surviving two fix attempts means stop: write the finding to your lessons
file, mark the item `[BLOCKED]`, and surface it. Don't thrash.

Files in this skill

  • README.md758 B
  • SKILL.md2.9 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.