fix(skills): sharpen parallel-ops routing per adversarial review (4 findings)

Codex refuter's 8-scenario cold-agent quiz found: (a) 'many files,
cheap models' matched both the native row and fleet-worker (fix:
discriminator is brain economics, not count); (c) 'overnight until X'
matched both loop-ops and iterate (fix: session shape, not duration —
tie-breaker note added); fleet-worker mislabeled 'same-provider only'
(it is provider-agnostic; the real split is one-brain-per-run vs mixed);
'every row above ends at fleet-ops' overclaimed (only branch-producing
rows). Also drops spawn's forbidden empty related-skills field.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

skills/parallel-ops/SKILL.md

@@ -20,20 +20,25 @@ you need, and stop reading here.
 
 | You want | Go to | Not this, because |
 |---|---|---|
-| One-off parallel subtasks, same provider, this session | native Workflow tool / `Agent` subagents (`isolation: worktree`) | not a fleet skill at all — no dedicated skill needed |
-| Cheap delegation — a few well-scoped tool-using subtasks on a cheaper brain | [fleet-worker](../fleet-worker/) | [fleetflow](../fleetflow/) is overkill for one brain |
-| Brains differ by work class, or you need cross-provider dissent in verify | [fleetflow](../fleetflow/) | [fleet-worker](../fleet-worker/) is same-provider only |
-| Recurring / scheduled / unattended loop | [loop-ops](../loop-ops/) | [iterate](../iterate/) is one session, not scheduled |
-| Drive ONE mechanical metric to a target, in one session | [iterate](../iterate/) | [loop-ops](../loop-ops/) is the scheduler *around* this |
-| Land/merge branches that parallel work produced | [fleet-ops](../fleet-ops/) | always the terminus — every row above ends here |
+| Parallel subtasks run by your session's OWN provider and model tier, in-process | native Workflow tool / `Agent` subagents (`isolation: worktree`) | not a fleet skill at all — no dedicated skill needed |
+| Work done by a CHEAPER brain than your session (GLM/Haiku/Sonnet-under-Opus) — one subtask or a whole fan-out, all one brain type | [fleet-worker](../fleet-worker/) | [fleetflow](../fleetflow/) is overkill when every worker runs the same brain |
+| DIFFERENT brains per work class in one run, or cross-provider dissent in verify (e.g. Codex refutes GLM) | [fleetflow](../fleetflow/) | [fleet-worker](../fleet-worker/) runs one brain type per run (any provider, but not mixed) |
+| Work that RECURS on a schedule across sessions — cron, routine, unattended ticks | [loop-ops](../loop-ops/) | [iterate](../iterate/) is one continuous session, not a schedule |
+| Drive ONE mechanical metric to a target in one continuous session (even a long overnight one) | [iterate](../iterate/) | [loop-ops](../loop-ops/) is the scheduler *around* sessions, not the session itself |
+| Land/merge branches that parallel work produced | [fleet-ops](../fleet-ops/) | the terminus for every branch-producing row above (in-process subagents and prompt authoring produce no branches) |
 | Author a static expert-agent prompt FILE (not a runtime worker) | [spawn](../spawn/) | listed only to catch the name collision with "spawn workers" |
 
+Tie-breakers for the two classic overlaps: "many files, cheap models" is
+**brain economics, not count** — cheaper brain → fleet-worker, own brain →
+native. "Run overnight until X" is **session shape, not duration** — one
+continuous run → iterate; scheduled re-entry across sessions → loop-ops.
+
 ## Two axes that confuse cold agents
 
 **Spawn vs. land.** fleet-worker and fleetflow *spawn* workers and produce
 branches; fleet-ops *lands* those branches through a test-gated queue.
-Every fleet lane ends at fleet-ops regardless of how it was spawned —
-agent team, background agent, `claude -p` worker, or human.
+Every branch-producing lane ends at fleet-ops regardless of how it was
+spawned — agent team, background agent, `claude -p` worker, or human.
 
 **Inner loop vs. outer loop.** iterate is the *inner* loop: one session,
 one metric, git as memory, runs until a stop condition. loop-ops is the