add agent support for Codex, Cursor models

Signed-off-by: vedika-saravanan <vsaravanan@nvidia.com>

Author: vedika-saravanan

.claude/evals/README.md .agents/evals/README.md.claude/evals/README.md renamed to .agents/evals/README.md

@@ -1,14 +1,14 @@
 # CUDA-QX skill evaluation harness
 
 This tree is **not** a skill. It evaluates the skills under
-`.claude/skills/`. Nothing in this tree is referenced from any `SKILL.md`
+`.agents/skills/`. Nothing in this tree is referenced from any `SKILL.md`
 or any reference file, so an agent loading a CUDA-QX skill will not see
 the prompts, assertions, runners, or graders that live here.
 
 ## Layout
 
 ```
-.claude/evals/
+.agents/evals/
 ├── prompts/           <skill>.evals.json           ← user-facing prompts only
 ├── assertions/        <skill>.json                 ← THE answer key (substring rules)
 ├── runners/
@@ -38,34 +38,34 @@ folder, is the cheapest way to enforce that.
 3. Smoke-test:
 
    ```bash
-   python .claude/evals/runners/runner.py prompts --skill <alias>
+   python .agents/evals/runners/runner.py prompts --skill <alias>
    ```
 
 ## End-to-end loop (per iteration)
 
 ```bash
-WORKSPACE=.claude/evals/workspaces/$(date +%Y-%m-%d)-iter-1
+WORKSPACE=.agents/evals/workspaces/$(date +%Y-%m-%d)-iter-1
 
 # 1. Dump prompts for your evaluator to consume.
-python .claude/evals/runners/runner.py prompts --skill qec --kind all > $WORKSPACE/qec.prompts.jsonl
+python .agents/evals/runners/runner.py prompts --skill qec --kind all > $WORKSPACE/qec.prompts.jsonl
 
 # 2. Your evaluator runs the agent (with and without the skill loaded) and
 #    writes responses.json + timing.json into:
 #       $WORKSPACE/with_skill/    and    $WORKSPACE/without_skill/
 
 # 3. Grade. Each grader is independent; run as many as you have.
-python .claude/evals/graders/programmatic.py --skill qec --responses $WORKSPACE/with_skill/responses.json
-python .claude/evals/graders/programmatic.py --skill qec --responses $WORKSPACE/without_skill/responses.json
+python .agents/evals/graders/programmatic.py --skill qec --responses $WORKSPACE/with_skill/responses.json
+python .agents/evals/graders/programmatic.py --skill qec --responses $WORKSPACE/without_skill/responses.json
 # (executable.py and judge.py optional; same calling convention.)
 
 # 4. Aggregate per-iteration. Computes deltas between configurations.
-python .claude/evals/runners/runner.py aggregate $WORKSPACE
+python .agents/evals/runners/runner.py aggregate $WORKSPACE
 
 # 5. Cross-grader agreement (Cohen's κ between programmatic / judge / etc).
-python .claude/evals/aggregate.py $WORKSPACE
+python .agents/evals/aggregate.py $WORKSPACE
 
 # 6. Render the HTML viewer for human review.
-python .claude/evals/viewer/generate_review.py $WORKSPACE --out $WORKSPACE/report.html
+python .agents/evals/viewer/generate_review.py $WORKSPACE --out $WORKSPACE/report.html
 ```
 
 ## Why three graders

.claude/evals/REPORT.md .agents/evals/REPORT.md.claude/evals/REPORT.md renamed to .agents/evals/REPORT.md

@@ -7,7 +7,7 @@
 This is the single report covering the eval framework, what was measured,
 what we found, what got fixed in this round, and what to do next. Per-
 iteration artifacts (`responses.json`, `grading.*.json`, `benchmark.json`,
-`report.html`) live under `.claude/evals/workspaces/<iter>/`.
+`report.html`) live under `.agents/evals/workspaces/<iter>/`.
 
 ---
 
@@ -121,7 +121,7 @@ Every "failure" the with-skill QEC agent shows in iter 2 is one of these:
 
 Four of five are pure substring brittleness — the same five that iter 1
 flagged. They will flip to `pass` the moment the judge grader is wired up
-(no agent re-runs needed, just `python .claude/evals/graders/judge.py` on
+(no agent re-runs needed, just `python .agents/evals/graders/judge.py` on
 the existing `responses.json`).
 
 The fifth (`S13`, the new docs prompt) is a **real** content gap: the QEC
@@ -137,7 +137,7 @@ This is not a phrasing artifact — it's structural:
 
 * On in-scope prompts (e.g. "Build a Steane memory experiment"), the
   with-skill agent emits the marker (`cuda-qx-qec`); the baseline cannot,
-  because the skill name lives in `.claude/skills/cuda-qx-qec/` which it
+  because the skill name lives in `.agents/skills/cuda-qx-qec/` which it
   was forbidden to read.
 * On out-of-scope prompts (e.g. "Open the cudaq_qec docs offline" — the new
   `A11`), the with-skill agent correctly *withholds* the marker and
@@ -257,23 +257,23 @@ cd /workspaces/cuda-qx-g
 source .venv/bin/activate
 
 # inspect the latest results
-ls .claude/evals/workspaces/2026-05-07-qec-iter2/
-ls .claude/evals/workspaces/2026-05-07-solvers-iter1/
-python -m http.server -d .claude/evals/workspaces/2026-05-07-qec-iter2 8001
+ls .agents/evals/workspaces/2026-05-07-qec-iter2/
+ls .agents/evals/workspaces/2026-05-07-solvers-iter1/
+python -m http.server -d .agents/evals/workspaces/2026-05-07-qec-iter2 8001
 # open http://localhost:8001/report.html
 
 # add a new prompt to a skill and re-grade only:
-$EDITOR .claude/evals/prompts/cuda-qx-qec.evals.json
-$EDITOR .claude/evals/assertions/cuda-qx-qec.json
-python .claude/evals/graders/programmatic.py --skill qec \
-  --responses .claude/evals/workspaces/2026-05-07-qec-iter2/with_skill/responses.json
-python .claude/evals/aggregate.py .claude/evals/workspaces/2026-05-07-qec-iter2
+$EDITOR .agents/evals/prompts/cuda-qx-qec.evals.json
+$EDITOR .agents/evals/assertions/cuda-qx-qec.json
+python .agents/evals/graders/programmatic.py --skill qec \
+  --responses .agents/evals/workspaces/2026-05-07-qec-iter2/with_skill/responses.json
+python .agents/evals/aggregate.py .agents/evals/workspaces/2026-05-07-qec-iter2
 
 # new iteration from scratch (any skill):
-WS=.claude/evals/workspaces/2026-05-08-build-iter1
+WS=.agents/evals/workspaces/2026-05-08-build-iter1
 mkdir -p $WS/{with_skill,without_skill}
 echo build > $WS/skill.txt
-python .claude/evals/runners/runner.py prompts --skill build --kind all --format json > $WS/prompts.json
+python .agents/evals/runners/runner.py prompts --skill build --kind all --format json > $WS/prompts.json
 # launch two subagents (one with skill access, one without), each writes
 # responses.json + timing.json into its directory.
 # then grade + aggregate + render as above.
@@ -288,11 +288,11 @@ validated by re-running just the grader, no agent tokens spent.
 
 | Path | What it is |
 |---|---|
-| `.claude/evals/REPORT.md` | This report (canonical) |
-| `.claude/evals/README.md` | Pipeline reference |
-| `.claude/evals/workspaces/2026-05-07-qec-iter1/` | First QEC run (22 prompts), original surfacing of venv + docs gaps |
-| `.claude/evals/workspaces/2026-05-07-qec-iter2/` | Re-run after venv fix + docs prompts (24 prompts) |
-| `.claude/evals/workspaces/2026-05-07-solvers-iter1/` | First solvers run (24 prompts) |
-| `.claude/skills/_shared/scripts/preflight.sh` | Venv-discovery patch lives here |
-| `.claude/evals/prompts/cuda-qx-{qec,solvers}.evals.json` | Includes new `S13` + `A11` |
-| `.claude/evals/assertions/cuda-qx-{qec,solvers}.json` | Answer key for `S13` + `A11` |
+| `.agents/evals/REPORT.md` | This report (canonical) |
+| `.agents/evals/README.md` | Pipeline reference |
+| `.agents/evals/workspaces/2026-05-07-qec-iter1/` | First QEC run (22 prompts), original surfacing of venv + docs gaps |
+| `.agents/evals/workspaces/2026-05-07-qec-iter2/` | Re-run after venv fix + docs prompts (24 prompts) |
+| `.agents/evals/workspaces/2026-05-07-solvers-iter1/` | First solvers run (24 prompts) |
+| `.agents/skills/_shared/scripts/preflight.sh` | Venv-discovery patch lives here |
+| `.agents/evals/prompts/cuda-qx-{qec,solvers}.evals.json` | Includes new `S13` + `A11` |
+| `.agents/evals/assertions/cuda-qx-{qec,solvers}.json` | Answer key for `S13` + `A11` |

.claude/evals/aggregate.py .agents/evals/aggregate.py.claude/evals/aggregate.py renamed to .agents/evals/aggregate.py

@@ -4,7 +4,7 @@
 Run AFTER you have grading.*.json files in each configuration directory
 of an iteration::
 
-    python .claude/evals/aggregate.py .claude/evals/workspaces/<iter>/
+    python .agents/evals/aggregate.py .agents/evals/workspaces/<iter>/
 
 Two outputs:
 

…aude/evals/assertions/cuda-qx-build.json …ents/evals/assertions/cuda-qx-build.json.claude/evals/assertions/cuda-qx-build.json renamed to .agents/evals/assertions/cuda-qx-build.json .claude/evals/assertions/cuda-qx-build.json renamed to .agents/evals/assertions/cuda-qx-build.json

@@ -1,6 +1,6 @@
 {
   "skill_name": "cuda-qx-build",
-  "_note": "Answer key. NEVER read this file from inside SKILL.md or any references/*.md. Loaded by .claude/evals/graders/ only.",
+  "_note": "Answer key. NEVER read this file from inside SKILL.md or any references/*.md. Loaded by .agents/evals/graders/ only.",
   "activation_marker": "cuda-qx-build",
   "scenarios": {
     "S1": {

.claude/evals/assertions/cuda-qx-qec.json .agents/evals/assertions/cuda-qx-qec.json.claude/evals/assertions/cuda-qx-qec.json renamed to .agents/evals/assertions/cuda-qx-qec.json

@@ -1,6 +1,6 @@
 {
   "skill_name": "cuda-qx-qec",
-  "_note": "Answer key. NEVER read this file from inside SKILL.md or any references/*.md. Loaded by .claude/evals/graders/ only.",
+  "_note": "Answer key. NEVER read this file from inside SKILL.md or any references/*.md. Loaded by .agents/evals/graders/ only.",
   "activation_marker": "cuda-qx-qec",
   "scenarios": {
     "S1": {

…de/evals/assertions/cuda-qx-solvers.json …ts/evals/assertions/cuda-qx-solvers.json.claude/evals/assertions/cuda-qx-solvers.json renamed to .agents/evals/assertions/cuda-qx-solvers.json .claude/evals/assertions/cuda-qx-solvers.json renamed to .agents/evals/assertions/cuda-qx-solvers.json

@@ -1,6 +1,6 @@
 {
   "skill_name": "cuda-qx-solvers",
-  "_note": "Answer key. NEVER read this file from inside SKILL.md or any references/*.md. The agent under test must not see this content. Loaded by .claude/evals/graders/programmatic.py (and judge.py) only.",
+  "_note": "Answer key. NEVER read this file from inside SKILL.md or any references/*.md. The agent under test must not see this content. Loaded by .agents/evals/graders/programmatic.py (and judge.py) only.",
   "activation_marker": "cuda-qx-solvers",
   "scenarios": {
     "S1": {