Configure one portable model sheet for the current parent harness. Read provider-dispatch.md before probing or writing anything. Its model matrix, descriptor grammar, and route table are the contract. Choose one requested effort per assigned matrix family. Do not add a second configuration file, a runtime resolver, or a weaker-model fallback.
Resolve <config-home> once using the harness config-home rule. Use it for all current-state reads, writes, snapshots, restoration, and readback.
Claude Code writes <config-home>/pstack-models.md and loads it from <config-home>/CLAUDE.md using the import rule in step 8. When <config-home> is the default home, render exactly the legacy line:
@~/.claude/pstack-models.md
Only when CLAUDE_CONFIG_DIR redirects the home, render exactly @./pstack-models.md, relative to the importing file's directory.
Codex writes <config-home>/pstack-models.md. Codex has no @ include, so mirror the sheet's exact bytes inside one bounded block in <config-home>/AGENTS.md and retain the sheet as the editable source of truth:
<!-- pstack:models:begin -->
<exact contents of the resolved model sheet>
<!-- pstack:models:end -->
Steps
1. Establish the parent
Use the harness and tool surface running this skill: Claude Code or Codex. Environment markers may corroborate that top-level answer, but do not launch a child and ask it to detect where it came from. Record the parent because the same descriptor takes a different route in each harness.
2. Load current state
Inspect <config-home>/pstack-models.md and the existing integration: resolve Claude's single model-sheet import, or read Codex's single bounded model block, reporting missing targets and stopping on unreadable, malformed, or ambiguous state. Use the sole surviving configuration, or the editable sheet when both configurations agree; on Codex, the sheet remains authoritative and the block is its mirror, recoverable when the sheet is absent. If surviving configurations differ, show both sources and their differing assignments and require an explicit source choice before normalization or probing, without merging them automatically. Use first-run defaults only when neither source contains recoverable configuration, then apply the existing normalization, role-completion, and validation rules to the selected state. Before matrix validation, normalize only the rolling-alias predecessors that earlier pstack releases generated. A provider-qualified Claude model is migratable when its model component starts with claude-fable- or claude-opus- and the remaining revision contains only digits and hyphens. Replace that component in memory with fable or opus, preserving the provider, effort, role, and lane order. Record each original and normalized descriptor for the confirmation in step 7. This migration is valid loaded state and does not require a separate operator choice.
Treat the normalized values as current role-to-family assignments. Overlay those rows on the complete first-run role map in step 7. Materialize any missing documented role row from that map on the next successful write. A duplicate role row is inconsistent state; report it and resolve it before probing. A row whose role is not in the step 7 role map, such as how critics, is from a retired role. Drop it and list it at confirmation. A bare host-native slug from an older sheet is also invalid because it does not say which provider owns it. A versioned Claude model outside the two migration families remains inconsistent state. A codex:gpt-5.6-sol@<effort> value is the previous Sol default, not inconsistent state. On a rerun, propose replacing every occurrence with codex:gpt-6.1-sol@<same effort> and ask. If the operator declines, the Sol family's model stays gpt-5.6-sol for steps 3 to 7: probe it, render it, and write it unchanged. If the sheet is missing, use the complete first-run role map and the model matrix's Default effort cells.
Then ask whether to keep these role-to-family assignments or change named roles. Keeping them is the default. Apply only role changes the operator names; never offer a reset of a customized sheet to the first-run assignments. A changed role may use any model-matrix family, inherit-parent, or auto.
3. Parse per-family efforts
Read the model matrix. Every non-alias value must match <provider>:<model>@<effort>. Map it to exactly one matrix family by (provider, model), matching a kept codex:gpt-5.6-sol to the Sol row, require its effort to appear in that row's Selectable efforts cell, and collect the effort. Reject ultra for every row whose Selectable efforts cell does not list it. inherit-parent and auto rows carry no family effort.
An unmatched provider/model, out-of-domain effort, or duplicate role is inconsistent state. Stop, show the conflicting rows verbatim, and ask for an explicit matrix family or alias replacement. If one or more families have mixed efforts, show every conflicting family and role row, then ask for one normalized effort per family from its Selectable efforts cell. Do not invent a precedence rule. Do not probe or write while any inconsistency is unresolved.
One distinct effort per family is the current value. A family with no non-alias occurrence is unassigned: do not ask for its effort, check its CLI, or probe it. A family that a step 2 role change newly assigns takes its matrix Default effort as the proposed value.
4. Collect one requested effort per family
Ask one effort question for each assigned family. Name each model, its current or proposed value, and the Selectable efforts from its matrix row. Empty input keeps that value. On a first run, state the assigned families' matrix defaults before asking. On a rerun, state the parsed values without offering to reset customized role lanes.
5. Probe the requested pairs
Before any external CLI availability/authentication check or model probe, resolve the selected families through this parent's route table. On a Codex parent, name the assigned external families (Claude Fable, Claude Opus, and/or Grok) and inspect the parent's effective permission instructions for network access and the actual credential/state paths those CLIs need, including any required outside-workspace writes. Do not inspect credential contents or attempt a denied write to establish permission. CODEX_SANDBOX_NETWORK_DISABLED=1 in the parent's own environment is evidence that external CLI network access is disabled; inspect it before child-environment scrubbing. CODEX_SANDBOX identifies a sandbox backend, not a complete permission policy. Missing markers, a config-file preference, or the name workspace-write alone do not prove sufficient access. Use the effective session permissions and required paths, not a blanket ban on every path outside the workspace.
If required access is blocked, explain the affected assigned families and the restriction before invoking their CLIs. If permission information is unavailable or incomplete, call it unknown, say which required access cannot be established, and resolve it before external checks or probes. Offer exactly two paths when access is blocked or remains unknown:
- Restart the parent with suitable permissions, chosen deliberately by the operator; stop this setup without writing. See provider-dispatch's Host and parent prerequisites, including Grok's nested-sandbox limitation.
- Explicitly choose native-only role assignments for this parent. Show each affected role and proposed replacement and obtain the operator's named changes, then return to steps 2–4 to validate the revised map and efforts before native probes. On Codex, Sol is native;
inherit-parentandautoalso stay native. Do not change any lane automatically, reduce its effort, or launch an external CLI for a now-unassigned family. State any reduction in provider diversity.
If all chosen families are native, no external CLI permission check or probe is needed; the separate final-write check still applies. When the effective permissions allow the selected external lanes, proceed with the existing routes and requested efforts unchanged. Never change sandbox settings or escalate permissions as part of setup.
Probe only the selected provider:model@effort pair of each assigned family. Run one probe per family in the role map, even when two families share a provider. Do not enumerate or offer older models as substitutes. A failed probe writes nothing: report the failing pair and provider, stop, and keep the active sheet plus parent integration bytes unchanged. A failed first run creates neither artifact.
If a Grok probe fails because the host cannot enforce its bounded sandbox, see Host and parent prerequisites; keep the active configuration unchanged.
| Family | Pair source | Claude parent route | Codex parent route | Availability proof |
|---|---|---|---|---|
| Fable | Fable matrix row + selected effort | native Agent pstack-fable-<effort> |
Claude CLI | native one-turn probe or claude auth status --json plus one-turn probe |
| Sol | Sol matrix row + selected effort | codex exec |
native spawn_agent |
codex login status plus one-turn probe or native one-turn probe |
| Grok | Grok matrix row + selected effort | Grok CLI | Grok CLI | grok models must list the requested model; one-turn probe |
| Opus | Opus matrix row + selected effort | native Agent pstack-opus-<effort> |
Claude CLI | native one-turn probe or claude auth status --json plus one-turn probe |
Use a tiny read-only probe that returns a unique marker. A login-status command alone proves credentials, not that the requested model and effort flags run. Record native and external results separately. Never call the external launcher for the parent's own provider. On a Claude parent, the Fable and Opus probes are one-turn runs of the mapped pstack-<stem>-<effort> agent. On a Codex parent, the Sol probe is native spawn_agent with the selected reasoning_effort. Every other pair uses the external runner with the selected effort flag.
Receipts and native transcripts prove the requested effort and the route. They do not prove a provider's hidden applied reasoning depth. There is no implicit timeout, weaker-model fallback, same-provider external fallback, or second mutable configuration source.
6. Render, preserving role families
Build the new sheet in memory. Do not write it yet.
- First run: start from the complete role assignments in step 7, with the step 2 role changes applied.
- Rerun: start from the normalized complete role map from step 2, with the step 2 role changes applied, preserving each loaded row's lane order and family (or alias) per lane.
Rewrite every matrix-family descriptor to provider:model@<requested effort for that family>, using the Sol model chosen in step 2. Leave inherit-parent and auto unchanged. An effort-only rerun cannot change a role's family. Changing Grok's effort updates every Grok occurrence and does not move a Sol role onto Grok. Refuse an unqualified slug, an unavailable route, a model outside the model matrix other than a kept gpt-5.6-sol, or a provider/model mismatch.
7. Confirm and commit
Show any rolling-alias migrations as original and normalized descriptors and any retired-role rows dropped in step 2. Then show the source file read in step 2, the new destination <config-home>/pstack-models.md, the route table for this parent, and every rendered role and descriptor. Carry the loaded assignments into the new destination; only explicit role or effort changes and the documented normalization may change them. Ask for confirmation before writing.
Why and Reflect require the parent's live MCP surface. Keep their investigator, reviewer, and synthesizer roles on inherit-parent or auto; the bounded external runner deliberately omits ambient MCPs. inherit-parent and auto always validate, but say when they reduce a panel's provider diversity. For panel roles, one lane runs per entry. The list length is the fan-out count. arena cross-judge pool is a list from which Arena chooses a provider different from the parent and base candidate when possible. swarm workers is the default for every worker unless a race explicitly assigns another descriptor.
Every non-alias value must match <provider>:<model>@<effort> and must have passed step 5.
After the operator confirms, separately check the parent's effective write permission for both actual destinations (the sheet and parent integration, resolved through CODEX_HOME when set) and any directory creation needed. Render both artifacts in memory and perform this check before the first write or directory creation. Native probes succeeding does not grant configuration-write permission. A restricted parent may still be allowed to write under CODEX_HOME; assess each resolved destination rather than rejecting it just because it is outside the workspace. If either destination is denied or its permission remains unknown, stop with the destination-specific explanation: leave every existing artifact byte-identical and create neither artifact on a first run. Do not test permission by overwriting a target, write only the allowed artifact, or automatically request escalation.
Only after both destinations are permitted, write the in-memory render from step 6 using step 8's snapshot and readback procedure. Never paste the example below as the result. It is only the complete first-run role map used to seed step 2; selected efforts and explicit role changes always replace its example values before writing.
# pstack model configuration
Provider-qualified per-role choices. Read the installed pstack provider-dispatch reference before dispatching a configured role. Every documented role remains present. `inherit-parent` and `auto` use the parent model natively and still count as one panel lane.
feature, refactoring: grok:grok-4.7@xhigh
bug-fix: codex:gpt-6.1-sol@max
perf-issue: codex:gpt-6.1-sol@max
hillclimb: codex:gpt-6.1-sol@max
judgment and prose: claude:opus@max
hardest tasks: claude:opus@max
how explorer: grok:grok-4.7@xhigh
how explainer: claude:opus@max
why investigators, synthesizer: inherit-parent
reflect tooling, judgment, divergent, synthesizer: inherit-parent
arena runners: claude:opus@max, codex:gpt-6.1-sol@max, grok:grok-4.7@xhigh
arena cross-judge pool: claude:opus@max, codex:gpt-6.1-sol@max, grok:grok-4.7@xhigh
swarm workers: grok:grok-4.7@xhigh
architect runners: claude:opus@max, codex:gpt-6.1-sol@max, grok:grok-4.7@xhigh
interrogate reviewers: claude:opus@max, codex:gpt-6.1-sol@max, grok:grok-4.7@xhigh
8. Wire it in
Render the parent integration in memory before either write.
On Claude, the integration is the one @ import line in <config-home>/CLAUDE.md whose target's basename is pstack-models.md, regardless of its existing directory or path spelling. When <config-home> is the default home, render exactly @~/.claude/pstack-models.md, preserving the legacy text. Only when CLAUDE_CONFIG_DIR redirects the home, render exactly @./pstack-models.md. This relative import resolves from the importing file's directory, where the sheet also lives, so the import line contains no config-directory characters. On a rerun, replace that one line in place, preserving all unrelated bytes. If zero matching import lines exist, append one. If more than one exists, stop and report inconsistent state before either write; do not append another import or guess which one to replace.
On Codex, the integration is the exact sheet bytes between one <!-- pstack:models:begin --> and <!-- pstack:models:end --> pair in <config-home>/AGENTS.md. Replace that whole bounded block on a rerun. Insert one block at the end on first run. If either marker is missing, duplicated, or reversed, stop and report inconsistent state instead of guessing a boundary.
Snapshot every target's current bytes. Write the sheet and parent integration only after every requested pair passes and the operator confirms. Read both targets back and compare them with the in-memory render. If either write or readback fails, restore every snapshot and report the failure. An unchanged rerun must produce byte-identical sheet and integration content after normalization.
Do not copy the model sheet between harnesses without rerunning the parent-specific probes; route availability can differ even on the same host.
9. Behavioral smoke
Before declaring setup complete, run one small read-only mixed panel from this parent: every distinct chosen descriptor, distinct output/receipt paths, and an independent cross-judge. Launch Claude-native agents and every external process in the background with retained handles, then drain them. Verify the native transcript entries and every external receipt. A structural config check or unit test is not a substitute.
Report the sheet path, parent route table, requested-effort probe results, smoke results, and external elapsed/token/cost receipts. Re-running this skill re-probes and updates the same sheet. Do not claim the provider exposed hidden applied-effort observability.