Issues
If a scanner finds something worth reporting, it creates an issue. An issue provides a detailed description of the finding along with supporting evidence.
List unresolved issues:
To view issue details, run:
Each issue provides a description of the problem, any applicable evidence it found, and instructions or guidance for resolving the issue.
Issues are written for both humans and Claude Code. You can resolve the issue yourself and close it by running:
This marks the issue completed. If you want to skip the issue (i.e. mark it as
a non-issue) use the --skipped option.
For more issue related commands, run gage issue -h.
It’s handy to use Claude to resolve issues. To instruct Claude to review open
issues and help you resolve them, use the /gage:resolve command, which is
installed with the Gage plugin.
Claude has several Gage tools for resolving issues, which will require your permission to use.
Authors and duplicate policies
Like notes, every issue records an author, and a plain write always inserts —
nothing constrains (name, author). A writer that wants to fold a re-scan into
its prior issue states a merge policy on the write (keep_status,
open_on_new_evidence, open_on_changed_evidence), which acts on the most
recent issue it wrote with the same name: new evidence is added, and the reopen
policies reopen a closed issue when the evidence warrants it. Author values
follow the scheme described in Notes: scanner:<name> for
deterministic writers, agent:...?call=<toolUseId> values tying model writers
to the authoring transcript entry, and user:<username> for people.