Agent skills

Claude Code Skills are packaged instructions an agent loads when a task matches what the skill covers. This repository ships two, at its root, for two different jobs.

ER7 concepts and terminology (segments, fields, repetitions, components, paths, delimiters), which of er7, er7-redact, and serde-er7 a task needs, the round-trip and absent/empty/null rules that are easy to get wrong, and short recipes for parsing, querying, editing, batch input, and redaction.

It teaches a downstream project how to use these crates as a dependency. It is not the guidance this repository's own agents follow to change the crates themselves — that is the other skill.

Copy er7-skill/ into a project's own .claude/skills/ directory. Claude Code picks it up automatically the next time a task matches its description — parsing or building an HL7® v2 message in Rust, an ER7 or batch-file question, an HL7 path such as PID-5.1, or removing patient detail before a message is logged, shared, or committed.

The canonical file is er7-skill/SKILL.md; this page is a summary of it, not a copy — read the file itself for the full rule list and every recipe.

The other direction: an agent packaging of this repository's own AGENTS.md files, for a downstream fork or another project's tooling that wants this workspace's conventions as a loadable skill rather than a set of files to go read first. It indexes, rather than duplicates: spec-driven development (spec first, then tests, then code, then docs), the four checks, and the rules that are easy to violate by skipping straight to the code — never real patient data, never collapsing absent/empty/null, no dictionary or validator added to a crate that deliberately has neither.

The canonical file is er7-rust-maintainer-skill/SKILL.md — read it, or the workspace AGENTS.md it indexes, before changing this repository.