fix reasoning-effort carriers for codex and agy, which both reject --effort

README.md

@@ -270,7 +270,7 @@ Reviewers run **in the order listed**, and whatever you list is exactly what run
 
 **Per-reviewer iteration caps** (`~max=<n>` suffix): caps how many **review → fix → re-review cycles** that one reviewer runs. It is the per-entry form of `--review-iterations`, and unlike that flag it reaches every reviewer type — including `codex`/`agy`/`claude`/`grok`/`cursor` and `ollama`, whose caps are otherwise fixed at 3 — so a single run can budget each reviewer separately: `--review-with claude~max=2,ollama~max=1,codex~max=3`. `<n>` is a non-negative integer; `0` means "loop until clean", bounded by a 10-iteration safety guardrail. A reviewer that stops because it spent a cap *you* set reports `capped`, which counts as clean for the merge gate — as opposed to `guardrail`, which is what a *built-in* cap reports when it cuts off a reviewer that was still finding real problems, and which blocks the merge.
 
-**Per-reviewer reasoning effort** (`~effort=<level>` suffix): specifies the reasoning effort level (`low`, `medium`, `high`, `xhigh`, `max`) for that reviewer: `--review-with codex[gpt-5.6-luna]~effort=max~opt`, `--review-with claude~effort=high~max=2`, `--review-with cursor[gpt-5]~effort=max`. For Cursor the suffix is folded into `--model` as `[effort=<level>]` (the CLI has no `--effort` flag); pair it with a `cursor[<model>]` bracket or a saved `--review-models cursor=…` default so there is a model to attach the variant to.
+**Per-reviewer reasoning effort** (`~effort=<level>` suffix): specifies the reasoning effort level (`low`, `medium`, `high`, `xhigh`, `max`) for that reviewer: `--review-with codex[gpt-5.6-luna]~effort=max~opt`, `--review-with claude~effort=high~max=2`, `--review-with cursor[gpt-5]~effort=max`. Each reviewer receives it in the form its own CLI accepts — `--effort` is **not** universal. `claude` and `grok` take the flag; **codex** takes `-c model_reasoning_effort=<level>`; **Cursor** folds it into `--model` as `[effort=<level>]`, so pair that one with a `cursor[<model>]` bracket or a saved `--review-models cursor=…` default; and **agy** picks the matching model variant from whatever `agy models` lists. Where a reviewer offers no such control — or no level matching what you asked — the effort falls back to prompt guidance rather than failing the review.
 
 `~max` applies in `series` mode (the default). In `--review-mode parallel` each reviewer runs a single review-only pass and the orchestrator applies the union once, so there are no per-reviewer cycles to cap — `~max` is ignored there with a warning.
 

commands/do/rpr.md

@@ -41,7 +41,7 @@ Parse `$ARGUMENTS` for `--issues` / `--no-issues` / `--issues-label <name>`: whe
    Note whether **any** completed review exists (from a copilot bot, a human, or another bot) — call this `HAS_EXISTING_REVIEW` — and specifically whether a **completed** `copilot-pull-request-reviewer` review exists (a node in `reviews.nodes`, NOT merely a pending review request) — call this `HAS_COPILOT_REVIEW`. Track a Copilot review that is only **pending** (Copilot present in `reviewRequests.nodes[].requestedReviewer` with no completed Copilot review yet) separately as `COPILOT_REVIEW_PENDING` — a pending-only review must NOT set `HAS_COPILOT_REVIEW`, or the "completed review exists" branch below would fire and resolve threads before Copilot has posted anything. Then dispatch on `REVIEW_AGENTS`:
 
    - **If `REVIEW_AGENTS` is empty** (no `--review-with`, no saved default, or an explicit `none`): request **no** review — not Copilot, not anything else. Proceed straight to step 3 and resolve whatever unresolved threads the PR already carries.
-   - **If `REVIEW_AGENTS` contains a local CLI (`codex`/`agy`/`claude`/`grok`/`cursor`):** run the **local-agent review loop** (`lib/local-agent-review-loop.md`, referenced below) for each such agent against the PR branch, forwarding `REVIEWER_APPLIES`, that entry's resolved `{REVIEW_MODEL}` (bracket, else saved `review-models[slug]` default, else empty → built-in default — see the loop reference below), that entry's `{REVIEW_EFFORT}` (from its `~effort=<level>`; empty when it carried none — Cursor folds this into `--model`, the other CLIs take it as `--effort`), and that entry's `{MAX_ITERATIONS}` / `{MAX_EXPLICIT}` (from its `~max=<n>`; the built-in `3` / `false` when it carried none). This produces findings (and, in reviewer-applies mode, fixes) locally — it does **not** request a Copilot cloud review for those slugs. Then proceed to step 3 to fetch and resolve any pre-existing unresolved threads as well. (If `REVIEW_AGENTS` also contains `copilot`, additionally run the Copilot path below.)
+   - **If `REVIEW_AGENTS` contains a local CLI (`codex`/`agy`/`claude`/`grok`/`cursor`):** run the **local-agent review loop** (`lib/local-agent-review-loop.md`, referenced below) for each such agent against the PR branch, forwarding `REVIEWER_APPLIES`, that entry's resolved `{REVIEW_MODEL}` (bracket, else saved `review-models[slug]` default, else empty → built-in default — see the loop reference below), that entry's `{REVIEW_EFFORT}` (from its `~effort=<level>`; empty when it carried none — the loop maps it to each CLI's accepted carrier), and that entry's `{MAX_ITERATIONS}` / `{MAX_EXPLICIT}` (from its `~max=<n>`; the built-in `3` / `false` when it carried none). This produces findings (and, in reviewer-applies mode, fixes) locally — it does **not** request a Copilot cloud review for those slugs. Then proceed to step 3 to fetch and resolve any pre-existing unresolved threads as well. (If `REVIEW_AGENTS` also contains `copilot`, additionally run the Copilot path below.)
    - **If `REVIEW_AGENTS` contains `ollama`:** run the **Ollama review loop** (`lib/ollama-review-loop.md`, referenced below) for each `ollama` entry against the PR branch (which must be checked out locally — the loop reviews a local `git diff`), forwarding that entry's `{OLLAMA_MODEL}` (empty = auto-select), its `{OLLAMA_EFFORT}` (from its `~effort=<level>`; empty when it carried none), and its `{MAX_ITERATIONS}` / `{MAX_EXPLICIT}` (from its `~max=<n>`; the built-in `3` / `false` when it carried none). Like the local-CLI path it produces findings the orchestrator applies locally (Ollama is always review-only) and does **not** request a Copilot review. Then proceed to step 3 to resolve any pre-existing threads. (If `REVIEW_AGENTS` also contains `copilot`, additionally run the Copilot path below.)
    - **If `REVIEW_AGENTS` contains `copilot`** (only ever because you asked for it — typed or saved): a `copilot~max=<n>` cap applies to this path too — see the cap-accounting rule in step 8. rpr accepts no `--review-iterations`, so `~max` is the only budget signal here, and each Copilot round below counts against it.
      - **A completed Copilot review exists** (`HAS_COPILOT_REVIEW`): skip requesting a new one — proceed to step 3 to address its threads.
@@ -128,7 +128,7 @@ Parse `$ARGUMENTS` for `--issues` / `--no-issues` / `--issues-label <name>`: whe
 
 ## Local-Agent Review Loop (for `--review-with codex|agy|claude|grok|cursor`)
 
-When `REVIEW_AGENTS` names a local CLI, step 2 (and the step-8 re-request) runs that agent's review against the PR branch via the shared local-agent loop. Pass `{REVIEW_AGENT}`, `{REVIEWER_APPLIES}`, that entry's resolved `{REVIEW_MODEL}` (the `<agent>[<model>]` bracket if the token carried one, else the saved `review-models[slug]` default resolved above — project over global, else empty → the reviewer's built-in default), that entry's `{REVIEW_EFFORT}` (from its `~effort=<level>`; empty when unset — without it Cursor cannot fold effort into `--model` and the other CLIs cannot pass `--effort`), the PR branch (`headRefName`), the base branch (`baseRefName`), and the project `{BUILD_CMD}`. Forwarding `{REVIEW_MODEL}` is what makes `--review-with=codex[o3]` and a saved `review-models` default actually pin the model on rpr's local passes — without it those passes would silently run the CLI's default model. The loop verifies build + tests in the main thread before pushing; afterward, continue to step 3 to resolve any pre-existing threads.
+When `REVIEW_AGENTS` names a local CLI, step 2 (and the step-8 re-request) runs that agent's review against the PR branch via the shared local-agent loop. Pass `{REVIEW_AGENT}`, `{REVIEWER_APPLIES}`, that entry's resolved `{REVIEW_MODEL}` (the `<agent>[<model>]` bracket if the token carried one, else the saved `review-models[slug]` default resolved above — project over global, else empty → the reviewer's built-in default), that entry's `{REVIEW_EFFORT}` (from its `~effort=<level>`; empty when unset — the loop's effort-carrier table maps it to each CLI's accepted form), the PR branch (`headRefName`), the base branch (`baseRefName`), and the project `{BUILD_CMD}`. Forwarding `{REVIEW_MODEL}` is what makes `--review-with=codex[o3]` and a saved `review-models` default actually pin the model on rpr's local passes — without it those passes would silently run the CLI's default model. The loop verifies build + tests in the main thread before pushing; afterward, continue to step 3 to resolve any pre-existing threads.
 
 !`cat ~/.claude/lib/local-agent-review-loop.md`