Configuration and team packs

ryni check . needs no configuration. To customize it, add ryni.toml at the scan root. For a file input, the scan root is that file's parent. Ryni does not search ancestor directories or merge nested configuration files.

schema-version = 1

[lint]
select = ["all"]
ignore = ["skill-directory-name"]

[discovery]
exclude = ["vendor", "target", "**/generated/**"]
respect-ignore = true

Unknown keys, rule IDs, malformed globs and unsupported schema versions are execution errors. --config path/to/ryni.toml selects a configuration explicitly; its path is relative to the working directory. --isolated ignores all configuration and packs. These two options cannot be combined.

Selection and precedence

Settings resolve in this order, from lowest to highest priority:

1. Built-in defaults: all six stable rules.
2. Local packs, in their listed order.
3. The repository configuration.
4. Explicit CLI overrides.

Each provided option replaces the earlier value. Arrays replace rather than append; an empty array clears the earlier value. Unspecified options retain the earlier value. select sets enabled rules; ignore is applied afterward, even when selection came from a later layer. Use a repository ignore = [] to clear a pack's ignore list.

Rule selectors are exact IDs or all. all includes stable rules and, when preview is enabled, preview rules. Without explicit selection, enabling preview does not change the default set. Selecting a preview rule without enabling preview is an error. Ignoring a preview rule does not require preview.

If a selected skill rule needs metadata, malformed frontmatter produces a skill-frontmatter prerequisite finding even if that rule was not selected or was ignored. Dependent checks are skipped. Selecting only Markdown rules does not report skill metadata problems.

Inspect the effective settings with:

ryni settings .
ryni settings . --config config/ryni.toml --select skill-name
ryni rule
ryni rule markdown-local-link
ryni rule --output-format json

A local team pack

A pack contains versioned rule selections and declarative requirements. Commit or vendor it with the repositories that use it. For example, config/team.toml:

schema-version = 1
name = "example/team"
version = "1.0.0"
requires-ryni = ">=0.2.2, <1"

[lint]
preview = true
select = ["all"]
required-metadata = ["owner"]
required-sections = ["Usage"]
required-files = ["AGENTS.md"]

Then reference it from ryni.toml:

schema-version = 1
packs = [{ path = "config/team.toml", version = "1.0.0" }]

Pack paths resolve relative to the configuration file. Versions must be exact SemVer versions, not ranges, and must match the pack's declared version. requires-ryni is a SemVer compatibility requirement. Pack names must be nonblank and unique within a configuration. Packs cannot import other packs.

The version pin validates the declared version; it is not a content-integrity lock. Reproducibility also requires keeping the pack contents pinned in version control. There is no remote pack fetching or executable plugin loading.

Requirements are parameters of the corresponding preview rules. They have no effect unless those rules are selected:

Option	Rule	Meaning
required-metadata	skill-required-metadata	Top-level YAML fields that must contain nonblank strings.
required-sections	markdown-required-sections	Exact, case-sensitive heading text required in every scanned Markdown file. Inline emphasis and code are read as text; fenced examples do not count.
required-files	project-required-files	Regular files required relative to the scan root, independent of exclusions. Existing directories do not satisfy this requirement.

Lists must contain unique, nonblank entries. Required file paths must be portable relative paths without . or .. components, colons, backslashes or null bytes. The duplicate-name rule compares parsed skill names across the files included in the scan. A file-only scan cannot detect duplicates elsewhere in the repository.

Discovery

By default, hidden directories are included, nested symlinks are skipped, and ignore-file filtering is enabled. Linked Markdown targets may still resolve through symlinks, point outside the project, or refer to excluded files.

Exclusions match paths relative to the scan root and their ancestors. Excluding vendor excludes its descendants. Exclusions apply even to explicitly supplied files. A file-only scan uses the parent as root, so excluding that file uses its basename. Patterns use gitignore syntax: patterns containing a slash are relative to the scan root; patterns such as *.md match at any depth.

Ryni respects .gitignore, .ignore, .git/info/exclude, and global Git excludes, including applicable parent and nested ignore files. Git ignore rules apply inside Git repositories; .ignore also works outside them. Matching files are skipped even if tracked by Git. Global excludes can make coverage differ between machines. Explicit file inputs bypass ignore-file filtering but still honor Ryni exclusions. respect-ignore = false or --no-ignore disables ignore-file filtering; --respect-ignore enables it again when configuration has disabled it. --exclude replaces the configuration's exclusion list and may be repeated; it also applies with --no-ignore.

A scan continues after recoverable read, discovery or target-inspection errors. Its available findings are still reported, but an incomplete scan exits with code 2, even if it also has lint findings.