Diagnostics & the RUNE#### codes

Rune analyzes a Runefile — parsing, resolving imports, and running the semantic analyzer — before it runs anything. Every statically detectable problem is reported with a file:line:col location, a caret-underlined span, and a stable RUNE#### code. The same analysis powers three surfaces:

execution — rune <task> refuses to run a Runefile with error diagnostics;
rune analyze — reports diagnostics on demand, for CI or scripts;
the language server (rune lsp) — publishes them live as you type.

Because all three share one analysis service, they report identical diagnostics.

`rune analyze`

rune analyze                 # analyze the discovered Runefile
rune analyze path/to/Runefile
rune analyze --json          # machine-readable output

Human output is one line per diagnostic plus a summary:

Runefile:8:13: error[RUNE2001]: unknown task: missing
1 error, 0 warnings

Exit codes: 0 (no error diagnostics), 3 (error diagnostics present), 2 (no Runefile found or it could not be read). Warnings never change the exit code on their own.

The codes

The codes are a stable contract: each condition maps to exactly one code, and a code’s meaning never changes once shipped. You can rely on them in output filtering and CI.

Parser (RUNE1xxx) — errors

Code	Condition
`RUNE1001`	Unexpected token
`RUNE1002`	Invalid indentation
`RUNE1003`	Unterminated string
`RUNE1004`	Incomplete expression
`RUNE1005`	Malformed task declaration
`RUNE1006`	Malformed parameter constraint (`enum` without a value list, empty `enum()`, a non-string enum value, or a value list on a non-`enum` kind)

Semantic (RUNE2xxx): errors, except `RUNE2010`, `RUNE2016`, and `RUNE2017` (warnings)

Code	Condition
`RUNE2001`	Unknown dependency
`RUNE2002`	Duplicate task
`RUNE2003`	Dependency cycle (lists every task in the cycle)
`RUNE2004`	Undefined variable
`RUNE2005`	Wrong argument count
`RUNE2006`	Duplicate parameter
`RUNE2007`	Invalid attribute
`RUNE2008`	Invalid setting
`RUNE2009`	Invalid executor (reserved; executors are open-ended custom interpreters)
`RUNE2010`	Public task lacks documentation — warning, never gates execution or exit code
`RUNE2011`	Invalid failure hook (a `\|\|` hook targets — or reaches through its dependencies — an `agent`-executor task)
`RUNE2012`	Unknown parameter type (supported kinds: `string`, `number`, `boolean`, `path`, `raw`, `enum`)
`RUNE2013`	Invalid enum values (duplicate value in a parameter’s `enum(...)` list)
`RUNE2014`	A parameter’s literal default violates its own type constraint
`RUNE2015`	Invalid annotation attribute (`[param-doc]` names an unknown parameter, duplicate `[param-doc]` for one parameter, or more than one `[returns]` on a task)
`RUNE2016`	Annotated kind name shadows a task name — warning: an unspaced legacy header may have re-parsed as a type annotation; write `name: task` for a dependency
`RUNE2017`	A string or path parameter reaches a shell command unquoted: warning, never gates execution or exit code. Write `{{quote(p)}}` (or `{{quote_each(p)}}` for a variadic) outside any shell quotes, or declare `p:raw` when word splitting is intended

Project (RUNE3xxx) — errors

Code	Condition
`RUNE3001`	Unresolved import
`RUNE3002`	Import cycle
`RUNE3003`	Duplicate imported namespace
`RUNE3004`	Incompatible Rune version

Notes

RUNE2008 (invalid setting), RUNE2010 (undocumented public task), and RUNE2017 (unquoted parameter) are surfaced by rune analyze and the language server. They are not raised on the execution path, so an existing Runefile keeps running — the editor still flags them.
Dependency- and import-cycle diagnostics attach every participating task/file as a related location, which editors show as clickable jumps.