GitHub Issues
Own issue lifecycle operations, evidence-based label/type selection, and explicit read-only taxonomy proposals. Product planning and orchestration remain with the caller.
Select the work
- For exact issue reads or lifecycle operations, including caller-selected metadata, comments, attachments, and relationships, read lifecycle.md.
- For content-based label/type selection on an existing issue, read metadata-classification.md. The request selects preview or application; classification does not authorize label removal or taxonomy changes.
- Only for an explicit request to audit, design, recommend, or propose issue taxonomy, read taxonomy-proposals.md. This branch is read-only and requires a repository, not a target issue.
These are request-derived branches, not persisted modes. Exact metadata operations do not require classification. Read states.md when interpreting native dependency, classification, or taxonomy results.
Shared boundaries
Before contacting GitHub, read the shared
network execution and
GitHub CLI preflight contracts.
Use authenticated gh; binary uploads use only the bundled attachment
transport documented in the lifecycle branch.
Resolve the exact repository and target before mutation. Preserve existing user authority: explicit write instructions authorize their scope; a preview or recommendation alone does not. Normalize operations with the shared options. Classification may derive separate label and type operations from one authorized classification request.
Keep free-form provider fields file-backed. Independently read mutations back and reconcile uncertain effects before retrying. Report exact issue URLs, verified changes or proposals, and unresolved evidence; do not infer success from a write receipt.
Use $g:github-investigation for root-cause or disposition judgment and
$g:github-repository-triage for issue/PR queues. Classification and taxonomy
analysis require Investigation only when their evidence needs deeper source or
history analysis.