Repository navigation
feat(pstack): upgrade Sol routing and review defaults - #24
Conversation
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
The summary sentence at skills/autoreview/SKILL.md:314 still states the old "Astra at medium" default and contradicts the upgraded gpt-6.1-sol/xhigh default applied everywhere else in this PR.
Review effort: Balanced
Findings: 1
What changed in this PR
This PR upgrades the PStack model-routing defaults. It moves implementation/synthesis roles to gpt-6.1-sol (high), replaces the previous Astra judgment/prose/hardest-task/reflection roles with gpt-6-sol (high), and raises the Architect, Interrogate, and Codex Autoreview review roles to gpt-6.1-sol at xhigh. The changes are almost entirely to the skill documentation that drives Poteto-mode routing, with matching updates to the autoreview helper's default constants and self-test, plus a changelog entry. It fits the repo's pattern of keeping per-role model defaults mirrored across models.md, the setup template, each routed skill, the autoreview engine, and the changelog.
Changes:
- Route code roles to
gpt-6.1-solhigh and judgment/prose/reflection roles togpt-6-solhigh acrossmodels.md,setup-pstack, and the Poteto playbooks/skills. - Slim Arena/Architect/Interrogate panels (drop Astra and
gpt-5.6-sol) and set Architect/Interrogate/Autoreview togpt-6.1-solxhigh. - Update the
autoreviewexecutable defaults (codex=gpt-6.1-sol, thinkingxhigh), its self-test, and help text; add a changelog entry.
| File | Description |
|---|---|
| skills/poteto-mode/references/models.md | Core per-role default table; source of truth for routing |
| skills/setup-pstack/SKILL.md | Setup template mirrors the routing defaults |
| skills/poteto-mode/SKILL.md | Inline spawn defaults updated to Sol 6.1 / Sol 6 |
| skills/poteto-mode/references/poteto-agent.md | Subagent implementation model bumped to gpt-6.1-sol |
| skills/poteto-mode/playbooks/{feature,bug-fix,perf-issue,hillclimb,refactoring}.md | Delegate model default bumped to gpt-6.1-sol |
| skills/arena/SKILL.md | Runner and cross-judge pool defaults updated |
| skills/architect/SKILL.md | Runner default set to gpt-6.1-sol xhigh |
| skills/interrogate/SKILL.md | Reviewer table reduced to Sol 6.1 xhigh + Opus xhigh |
| skills/reflect/SKILL.md | Judgment/divergent/synthesis to Sol 6; tooling to Sol 6.1 |
| skills/how/SKILL.md, skills/why/SKILL.md | Explainer/synthesizer defaults bumped to gpt-6.1-sol |
| skills/codex-first/SKILL.md | Codex exec model default updated to gpt-6.1-sol |
| skills/autoreview/SKILL.md | Docs updated to Sol 6.1 xhigh, but one summary line (314) left stale |
| skills/autoreview/scripts/autoreview | Default constants + self-test + help text updated to Sol 6.1 xhigh |
| CHANGELOG.md | New entry describing the routing upgrade |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Upgrade implementation and synthesis roles to GPT-6.1 Sol high, and replace Astra judgment/reflection roles with GPT-6 Sol high. Arena now runs Sol 6.1, Opus 5.5, and Fable 5.1 at high; its judge pool is Sol 6.1 and Opus high.
Architect uses Sol 6.1 xhigh alongside Opus/Fable high. Interrogate uses one Sol 6.1 xhigh reviewer plus Opus xhigh. Codex Autoreview defaults to Sol 6.1 xhigh. Align setup templates, inline defaults, executable defaults, and changelog.
Validation: