punk-supervisor: no way to cancel a running task

[t3chn]

Problem

No mechanism to cancel a task after dispatch. Once claude -p is running, the only option is kill <pid> manually.

Expected

punk cancel <task-id>

Should: find the process (from slot tracking), send SIGTERM, wait 5s, SIGKILL if needed, move to done/ with status=cancelled, log cancellation event in audit.jsonl.

Context

During adjutant build, 3 parallel agents were dispatched. Wanted to cancel 2 after realizing they'd conflict, but had no mechanism.

[t3chn]

Specpunk issue admission

• policy: incident_admission_v1
• decision: close_now
• source_kind: unknown
• has_runtime_report: false
• has_deterministic_evidence: false
• has_runtime_marker: false
• duplicate_or_resolved: false
• high_severity_marker: false
• targets_legacy_surface: true
• targets_active_core_surface: false
• targets_later_track: true
• indicates_core_blocker: false
• targets_cut_surface: false

Matched runtime markers

• none

Reasons

• issue targets legacy/decommissioned surfaces: references decommissioned punk-supervisor surface, references legacy supervisor audit.jsonl surface

Only core_now issues are eligible for immediate core-stabilization work intake. defer_after_core stays open but outside the active loop. close_now is the close-now path.