feat(skills): /issue — file GitHub issues from arbitrary input

.claude/rules/tooling-reference.md

@@ -94,9 +94,10 @@ Subagents with `memory: project` in this setup: `scout`, `triage`, `planner`, `p
 
 Operator-only slash commands (`disable-model-invocation: true` — never auto-invoked) that sit on top of the pipeline. The cron `op-pipeline-orchestrator` keeps driving the pipeline forward in the background; these are the on-demand surface for operating it.
 
-- **`/run [--repo <name>] [--since <duration>] [--stage <name>] [--fix-mode]`** — drives the pipeline forward across in-scope items, dispatches the right subagent per stage, classifies every touched item, writes a structured report to `~/.claude/orchestrator-log/run-<UTC-timestamp>.md`. `/merge` and `/unblock` consume this report shape.
-- **`/merge [--risky] [--repo <name>]`** — read-only ranked merge queue. Reads the latest `/run` report (or derives via gh if stale), computes calibrated HIGH/MED/LOW confidence per PR, emits copy-paste merge commands. Never merges itself; the calibration is built so MED is the median outcome, not HIGH.
+- **`/run [--repo <name>] [--since <duration>] [--stage <name>] [--fix-mode]`** — drives the pipeline forward across in-scope items, dispatches the right subagent per stage, classifies every touched item, writes a structured report to `~/.claude/orchestrator-log/run-<UTC-timestamp>.md` where `<UTC-timestamp>` is `YYYY-MM-DDTHH-MM-SSZ` (ISO-8601 basic, UTC, colons replaced with hyphens for filename safety; e.g. `run-2026-04-25T11-00-38Z.md`). `/merge` and `/unblock` consume this report shape.
+- **`/merge [--risky] [--repo <name>]`** — read-only ranked merge queue. Reads the latest `/run` report (or derives via gh if stale), computes calibrated HIGH/MED/LOW confidence per PR, emits copy-paste merge commands. Never merges itself. Calibration specifics live in the `/merge` skill spec.
 - **`/unblock [--section credentials|decisions|sensitivity] [--repo <name>]`** — surfaces every pending decision only the operator can make: missing credentials, decision conflicts against `$OP_CONTEXT_REPO`, sensitivity escalations. Every row has a specific actionable next step. Sensitivity section never leaks private content per `context-repo.md` STAGE 13.
+- **`/issue <text> [--repo <name>] [--dry-run] [--no-confirm]`** — files one or many GitHub issues from arbitrary input (a sentence, paragraph, transcript, email). Auto-detects single-vs-batch: ≤3 sentences with a resolvable repo → Mode A (draft + confirm + file); longer or multi-item input → Mode B (decompose, classify each as NEW / DUPLICATE_OF / ALREADY_RESOLVED / OUT_OF_SCOPE, render approval table, file approved rows only); ambiguous short input → Mode C (one clarifying question, no second). Sanitizes every body against the target repo's sensitivity tier per `context-repo.md` (strict strip on public repos, mild on `staff-ok`, never paste verbatim transcript/email content). Refuses to run when `gh auth status` shows the bot account active — the bot triages, the operator files. Forbids `auto:approved` and `override:*` labels; sets only `stage:triaged` plus type/severity/component so the issue is drop-in acceptable to scout/triage.
 
 ## Context repo