Coding Agents

Coding agents read their instructions from files in the repository. Every agent reads the same files, so the rules are written once.

Instruction files

File	Holds	Read by
AGENTS.md	General coding guidelines	Codex, OpenCode, GitHub Copilot and most other agents
PROJECT.md	Build and test commands, verification, shipping, pull request checks, code standards and per-project conventions	Every agent, through AGENTS.md
CLAUDE.md	Imports AGENTS.md and PROJECT.md	Claude Code
.github/copilot-instructions.md	A pointer to AGENTS.md and PROJECT.md	GitHub Copilot

Change the rules in AGENTS.md or PROJECT.md; the other files only point to them.

Skills

Task-specific instructions are Agent Skills. Each skill is a directory .agents/skills/<name>/ holding a SKILL.md file and, where needed, scripts. Codex and OpenCode find skills in .agents/skills/.

Claude Code reads skills from .claude/skills/ only. Each skill therefore has a stub .claude/skills/<name>/SKILL.md with the same frontmatter, whose body sends the agent to the canonical file. A symbolic link would check out as a plain file on Windows clones without symlink support. The Lint Miscellaneous Scripts workflow fails when a stub's frontmatter differs from its canonical skill.

The repository provides these skills:

Skill	Use it to
upgrade-npm-packages	Upgrade npm dependencies behind a ten-day release cooldown; see Upgrade NPM Packages
capture-docs-screenshots	Retake the user documentation screenshots from a running installation
review-pull-request	Check that a pull request passes every quality gate before merge
check-code-quality	Review the code quality of one component
plan-before-coding	Analyse the codebase and requirements before implementing
architecture-design	Design an architecture and its diagrams and documentation
react-frontend-engineer	Work on the React client
playwright-tester	Explore the website and write Playwright tests
technical-debt-remediation-plan	Plan the remediation of technical debt
critical-thinking	Challenge assumptions behind a proposed solution
mentor	Guide an engineer rather than writing the code for them

Seven of these skills are adapted from awesome-copilot under the MIT licence; see .github/REFERENCES.md.

Adding a skill

1. Create .agents/skills/<name>/SKILL.md. Its frontmatter needs name, equal to the directory name, and a description that says what the skill does and when to use it. The agent decides whether to load the skill from the description alone.
2. Create .claude/skills/<name>/SKILL.md with the same frontmatter and a body that points to the canonical file.
3. Add the skill to the table above.
4. Validate both directories with skills-ref: skills-ref validate .agents/skills/<name>.

Using coding agents

Contributors use coding agents for co-development, code review and draft pull requests. Disclose code that a coding agent generated, particularly when it carries significant program logic. Understand every line you contribute: review generated code before submitting it, and keep the critical thinking for yourself.