Skip to content

feat(pstack): support GPT-6 role families - #9

Merged
thisguymartin merged 3 commits into
mainfrom
feat/astra-model
Sep 28, 2026
Merged

thisguymartin merged 3 commits into
mainfrom
feat/astra-model

Conversation

@thisguymartin

@thisguymartin thisguymartin commented Sep 28, 2026 •

Copy link
Copy Markdown
Owner

Closes #8

Why

setup-pstack rejects GPT-6 Astra, Sol, and Luna because its model matrices do not list them. Add them as optional Codex families so Architect and other roles can select them without changing existing assignments.

Scope

  • Add astra, sol-6, and luna to the additional model matrix with high as the proposed pstack effort.
  • Route Codex parents through native spawn_agent probes and Claude Code parents through the existing external Codex runner.
  • Check all five supported effort values and the runner's model and effort arguments. Keep the stock matrix and first-run role map unchanged.
  • Merge current main (multiple gateway model choices). The one conflict was in the flex-matrix test, and it resolves to main's version.
  • Document the families in docs/LANES.md, docs/USAGE.md, docs/reference.md, README.md, and UPSTREAM-FLEX.md. The docs cover the panel-diversity rule (all Codex families count as one provider) and why Codex's ultra effort is not selectable.

Blast Radius

Existing model sheets and default roles keep their current descriptors. Only roles that select a new GPT-6 family use the new rows.

Verification

  • Bun tests, strict typecheck, static invariants, and plugin validation pass. After the merge with main, the full suite reports 197 passing tests.
  • The exact candidate is installed in every affected harness.
  • The changed behavior passes from each real user surface.
  • The installed version, action, and observed result appear below.

Live evidence:

  • Codex parent route: native Codex probes for gpt-6-astra@high, gpt-6-sol@high, and gpt-6-luna@high returned their expected markers in an earlier session.
  • Claude Code parent route: from a Claude Code session, this candidate's pstack-runner --parent claude --provider codex --effort high --mode read-only ran once per model on Codex CLI 0.157.1. All three exited 0 with receipt status: complete, modelEvidence: pinned-argv, the exact expected marker, and elapsed times of 6734 ms (Astra), 6963 ms (Sol), and 5725 ms (Luna).

These probes prove model availability and both routes. They do not prove installed plugin behavior. Keep this PR in draft until /setup-pstack passes in both harnesses with the exact candidate installed.

Add Astra, GPT-6 Sol, and Luna as optional model families for setup-pstack. Keep upstream defaults intact and validate their provider routes and effort flags.
# Conflicts:
#	plugins/pstack/skills/poteto-mode/scripts/runner/model-matrix.test.ts
Add the astra, sol-6, and luna families to the lane guide, usage guide,
reference, README, and fork-ownership list. Existing sheets and first-run
defaults stay on codex:gpt-5.6-sol@max.
@thisguymartin
thisguymartin marked this pull request as ready for review September 28, 2026 06:23
@thisguymartin
thisguymartin merged commit 06e3e0e into main Sep 28, 2026
1 check passed
@thisguymartin
thisguymartin deleted the feat/astra-model branch September 28, 2026 17:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Support GPT-6 models for pstack role assignments

1 participant