Summary
pstack 0.15.0 (/setup-pstack and the skill defaults) hard-codes these Task model slugs:
grok-4.6-fast-xhigh
claude-fable-5-1-thinking-max
gpt-5.6-sol-max
claude-opus-5-thinking-xhigh
Spawning a subagent with any of them fails:
Invalid model selection "grok-4.6-fast-xhigh". Model could not be resolved to a valid subagent model.
Allowed model slugs:
- claude-fable-5-1-thinking-high
- claude-fable-5-thinking-high
- claude-opus-5-thinking-high
- composer-2.5-fast
- cursor-grok-4.6-high-fast
- gpt-5.6-luna-medium
- gpt-5.6-sol-medium
- gpt-5.6-terra-medium
Enabling Fable 5.1 / Sol / Opus 5 / Grok 4.6 in Settings → Models (and the Explore Subagent Model dropdown) does not add the -max / -xhigh / grok-4.6-fast-xhigh ids to that allow-list. /setup-pstack then either writes slugs that cannot spawn, or remaps to a different effort tier.
skills/interrogate/SKILL.md already says: if a slug is rejected, pick the closest Task-valid equivalent and open a PR to update the default table. This is that report.
Suggested pstack fix
Update the default table in skills/setup-pstack/SKILL.md (and the matching hardcoded defaults in poteto-mode, arena, architect, interrogate, how, why, swarm, reflect) to the current Task-valid slugs of the same families:
| current default |
spawnable equivalent |
grok-4.6-fast-xhigh |
cursor-grok-4.6-high-fast |
claude-fable-5-1-thinking-max |
claude-fable-5-1-thinking-high |
gpt-5.6-sol-max |
gpt-5.6-sol-medium |
claude-opus-5-thinking-xhigh |
claude-opus-5-thinking-high |
/setup-pstack already treats the session Task allow-list as source of truth. The baked-in defaults should match that list so a fresh install works before the user writes pstack-models.mdc.
Cursor-side gap (not fixable in this plugin alone)
Task still does not expose -max / -xhigh even when those models are selected in the picker. If that is intentional, pstack defaults should stay on the spawnable slugs. If it is a product bug, the allow-list needs to include the effort tiers pstack names.
Repro
- Enable Claude Fable 5.1, GPT-5.6 Sol, Claude Opus 5, Cursor Grok 4.6 in Settings → Models.
- Install pstack, run
/setup-pstack (or use the skill defaults).
- Have an agent
Task a subagent with model: grok-4.6-fast-xhigh (or any of the four defaults).
- Observe the invalid-model error and the allowed-slug list above.
Summary
pstack 0.15.0 (
/setup-pstackand the skill defaults) hard-codes theseTaskmodel slugs:grok-4.6-fast-xhighclaude-fable-5-1-thinking-maxgpt-5.6-sol-maxclaude-opus-5-thinking-xhighSpawning a subagent with any of them fails:
Enabling Fable 5.1 / Sol / Opus 5 / Grok 4.6 in Settings → Models (and the Explore Subagent Model dropdown) does not add the
-max/-xhigh/grok-4.6-fast-xhighids to that allow-list./setup-pstackthen either writes slugs that cannot spawn, or remaps to a different effort tier.skills/interrogate/SKILL.mdalready says: if a slug is rejected, pick the closest Task-valid equivalent and open a PR to update the default table. This is that report.Suggested pstack fix
Update the default table in
skills/setup-pstack/SKILL.md(and the matching hardcoded defaults in poteto-mode, arena, architect, interrogate, how, why, swarm, reflect) to the current Task-valid slugs of the same families:grok-4.6-fast-xhighcursor-grok-4.6-high-fastclaude-fable-5-1-thinking-maxclaude-fable-5-1-thinking-highgpt-5.6-sol-maxgpt-5.6-sol-mediumclaude-opus-5-thinking-xhighclaude-opus-5-thinking-high/setup-pstackalready treats the session Task allow-list as source of truth. The baked-in defaults should match that list so a fresh install works before the user writespstack-models.mdc.Cursor-side gap (not fixable in this plugin alone)
Task still does not expose
-max/-xhigheven when those models are selected in the picker. If that is intentional, pstack defaults should stay on the spawnable slugs. If it is a product bug, the allow-list needs to include the effort tiers pstack names.Repro
/setup-pstack(or use the skill defaults).Taska subagent withmodel: grok-4.6-fast-xhigh(or any of the four defaults).