Explicit issue administration

This skill handles identified issue fields and explicitly requested comments only. It is separate from the luna-submit skill: read-only lookups use a direct authenticated provider tool; authorized writes use one gpt-6-luna worker at xhigh while the parent keeps its own model settings.

Before a write

1. Resolve the provider, repository/project and exact issue identifier. Read the current issue and the provider's allowed fields/states.
2. Build a packet naming exactly one requested field or comment and its value, excluded fields, current state, repository rules, evidence requirements and delegation_depth: 0. Group related fields only when the user explicitly names them together. Do not infer a title, label, assignee, state or completion from CI, tests or unrelated issue activity.
3. For a description request, send only the description body. For a comment request, send only the exact comment body and no field changes. For title, label, assignee or status, send only that field. Preserve all other fields.
4. If an assignee name has multiple matches, show stable account identifiers and ask the parent for a choice. If a requested state is unavailable or ambiguous, show the provider's choices and ask. A missing connector or unsupported write is blocked; do not install tools or substitute a label.

Read-back and uncertain writes

Use one writer for an issue. After a write, read back only the requested field and the issue identifier. A matching field is completed; an inaccessible target with other targets confirmed is partial. A timeout or ambiguous result requires a read-back by content, author and timing before retry. If the original write cannot be reconciled, return uncertain and do not create a duplicate comment or issue. The worker must not spawn another submission worker.

Return the shared worker packet with the provider action, requested field or comment, read-back evidence, status and remaining question. Keep response fields bounded and preserve permission errors as permission errors; they do not prove an issue is absent. The default write effort is xhigh; a later low-effort comparison requires an explicit preference change and is never implicit.

Do not schedule monitors or infer completion. Return semantic diagnosis, provider-specific state mapping and unresolved choices to the parent.