| name | criv |
| description | Use when working in a criv vault, refreshing criv state, looking for undocumented code, or choosing which criv skill covers a task. |
criv
criv keeps repository documentation connected to source code. A vault is
green when criv check prints criv check: ok. Every criv workflow ends
green.
The loop
criv watch --once refreshes .criv/state.json after a code, docs, or ADR change.criv check reports drift.criv enforce --stage ci gates a change to ADR-governed source.
The checking-drift skill owns the flags and the failure paths.
Finding what is undocumented
criv query nodes --kind code --without-docs lists code that no document claims.criv query coverage --by module and criv query coverage --by adr report coverage.criv query --help holds the full query surface. Add --limit <N> to bound a
large answer, and --format ndjson to read one row per line.
Reading a failure
A criv check diagnostic always names a code and a repair. Other failures name
a code only when criv has a defined recovery for them; the rest report prose.
When a [code] or a fix: line is present, use it and run the command it
names. The checking-drift skill holds the codes, the machine-readable
formats, and the recovery paths.
Never guess a flag: read it from criv <command> --help, or from
criv --usage-json for the whole command tree.
Which skill covers the task
referencing-code — wiki-link syntax from a document or ADR to source.writing-decisions — ADR frontmatter, governs scopes, and policy patterns.checking-drift — the check loop, its flags, and how to read a diagnostic.c4-authoring — the LikeC4 architecture workspace.criv-me — developing a decision with the user before writing it down.
The work is complete when the change is written, criv watch --once has run
after it, and the vault is green.
