Update PullOps Go supervision rules for run state, leases, and stalls

[daKmoR]

What to build

Update the PullOps Go skill and its installed contract so an observing agent supervises long-running local runs through JSONL PullOps Progress Events plus PullOps Run State. PullOps Go should report semantic milestones immediately, issue compact healthy-run updates every 5-10 minutes, wait while a PullOps Lease is active, reconcile v1 PullOps Liveness Signals after lease expiry, record PullOps Stall Classification before intervention, and avoid artifact, process, git, CI, or GitHub probing while a run remains healthy.

Acceptance criteria

• PullOps Go instructions require supervision through JSONL PullOps Progress Events plus PullOps Run State during healthy execution, with PullOps Progress Events treated as semantic progress rather than heartbeat noise.
• The skill contract requires immediate reporting for milestones and compact healthy-run updates every 5-10 minutes without displaying every PullOps Heartbeat.
• The supervision rules require waiting while leases are active, reconciling advanced heartbeat or changed child run set after lease expiry, and recording PullOps Stall Classification before stopping, retrying, or replacing work.
• The instructions forbid killing unrelated processes, resetting or discarding local changes, starting parallel same-branch work before reconciliation, and using logs, git diff, CI, or GitHub state as required v1 liveness signals.
• Skill contract or setup-output tests assert the installed PullOps Go guidance includes JSONL plus run-state supervision, lease reconciliation, stall classification, healthy reporting cadence, and no healthy-run artifact or process probing.

Blocked by

• Add PullOps Run State and Heartbeat primitive #145
• Wire PRD Child Coordination into PullOps Run State #146

Covered PRD user stories

• 1
• 3
• 7
• 8
• 9
• 10
• 11
• 19
• 20
• 21
• 22
• 23
• 24
• 25

[daKmoR]

PullOps closed this Child Issue because PR #151 merged into the PRD branch pullops/prd-144.