aliasunder · Nov 11, 2026 · Nov 11, 2026 · Nov 11, 2026 · Nov 11, 2026 · Nov 12, 2026
diff --git a/AGENTS.md b/AGENTS.md
@@ -28,7 +28,7 @@ src/
   config.ts                # action inputs → validated ActionConfig
   logger.ts                # structured JSON logger — levels, child contexts, lazy props
   github/                  # GitHub I/O: event payload → PrContext, octokit wrappers (diff fetch, review posting)
-  openrouter/              # OpenRouter I/O: @openrouter/sdk wrapper, per-attempt and shared review deadlines, retry ladder (up to two attempts per model, then the fallback_model input; on HTTP errors, timeouts, invalid structured output), cost summary
+  openrouter/              # OpenRouter I/O: review requests through @openrouter/agent, per-attempt and shared review deadlines, retry ladder (up to two attempts per model, then the fallback_model input; on HTTP errors, timeouts, invalid structured output), cost summary
   diff/                    # pure transforms over parse-diff output: git-quoted path decoding, annotation, commentable lines, diff-level exclusion (patterns, gitattributes linguist rules, wildcard cap: at most 2 "*" per path segment so matching cannot backtrack for minutes)
   context/                 # workspace I/O: conventions file, root .gitattributes, changed files, import-trace scan, doc-mention scan, priority docs
   review/                  # pure review logic (the model call is a function the caller passes in): finding schema, phases + stage dispatch (defined under Review instruction authoring), prompt, non-finding filter and unknown-file filter (README sections of those names), cross-phase merge, path normalization, selection, comment mapping, markdown code spans, title similarity, context notes, summary
@@ -47,7 +47,9 @@ backwards dependency and a bug — enforced by a
 the outermost layer, so only tests import from a `__tests__/` folder.
 
 Use the official SDKs — don't hand-roll what `@actions/core`,
-`@actions/github` (octokit), `@openrouter/sdk`, or `parse-diff` already do.
+`@actions/github` (octokit), `@openrouter/agent`, `@openrouter/sdk` (a dev
+dependency: the tests take Responses result and error shapes from it), or
+`parse-diff` already do.
 Before writing any parsing or validation helper, check the SDK's utility
 surface first: `@actions/core` ships `getBooleanInput` (strict YAML 1.2
 booleans), which replaced a hand-rolled boolean-string Zod transform here.

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -43,7 +43,8 @@ points:
   constructs SDK clients and touches `process.env`, except that `logger.ts`
   reads `LOG_LEVEL`
 - Use the official SDKs (`@actions/core`, `@actions/github`,
-  `@openrouter/sdk`, `parse-diff`) — don't hand-roll what they already do
+  `@openrouter/agent`, `@openrouter/sdk` for Responses result and error shapes
+  in tests, `parse-diff`) — don't hand-roll what they already do
 - Relative imports carry explicit `.js` extensions (ESM runtime requirement)
 - Tests are behavioral specs: one focused `it()` per behavior, exact
   assertions, `const` per test

diff --git a/README.md b/README.md
@@ -81,7 +81,7 @@ An empty value, such as an unset repo variable, selects the input's default, so
 | `model`                      | `anthropic/claude-sonnet-4-6` | OpenRouter model slug exactly as listed on openrouter.ai/models                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            |
 | `fallback_model`             | `""`                          | Model tried after the primary model's attempts fail (up to two per model). A timed-out primary attempt moves to this model immediately, without retrying the primary. Auth and credit errors (HTTP 401/402/403) stop the review without trying it. Empty = no fallback                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     |
 | `request_timeout_seconds`    | `900`                         | Deadline for one model request, in seconds, capped by the time left in `review_timeout_seconds`. When it elapses, the attempt is recorded as `timeout` and the review moves to `fallback_model` if one is set; a timeout on the last model (the primary, when no fallback is set) may be retried once. The HTTP call is aborted best-effort: a request the provider keeps serving is still billed, and its cost-summary row shows `n/a`. Empty = 900                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       |
-| `review_timeout_seconds`     | `1500`                        | Shared model-work budget from action start, in seconds, including context preparation, all phases, retries, fallbacks and cost lookups. Completed phases post when it expires, with incomplete coverage identified; the run fails if none completed. Leave room below the workflow job timeout for setup and GitHub publication. External cancellation stops the run at once; unlike the deadline, it does not post completed phases. Empty = 1500                                                                                                                                                                                                                                                                                                                                                                                                                                                                                         |
+| `review_timeout_seconds`     | `1500`                        | Shared model-work budget from action start, in seconds, including context preparation, all phases, retries and fallbacks. Completed phases post when it expires, with incomplete coverage identified; the run fails if none completed. Leave room below the workflow job timeout for setup and GitHub publication. External cancellation stops the run at once; unlike the deadline, it does not post completed phases. Empty = 1500                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       |
 | `max_findings`               | `""` _(uncapped)_             | Cap on posted findings, highest severity first. Empty = all validated findings post.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       |
 | `severity_threshold`         | `low`                         | Minimum severity to post: `low` \| `medium` \| `high` \| `critical`                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        |
 | `conventions_file`           | `AGENTS.md`                   | Repo-relative path to the conventions file included in the prompt (truncated at `conventions_budget_tokens`). When the PR also changes the file, its full text is sent at most once, never twice                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           |

diff --git a/action.yml b/action.yml
@@ -25,7 +25,7 @@ inputs:
     required: false
     default: "900"
   review_timeout_seconds:
-    description: Shared model-work budget from action start, in seconds, including context preparation, all phases, retries, fallbacks and cost lookups. Completed phases post when it expires, with incomplete coverage identified; the run fails if none completed. Leave room below the workflow job timeout for setup and GitHub publication. External cancellation stops the run at once; unlike the deadline, it does not post completed phases. Empty = 1500
+    description: Shared model-work budget from action start, in seconds, including context preparation, all phases, retries and fallbacks. Completed phases post when it expires, with incomplete coverage identified; the run fails if none completed. Leave room below the workflow job timeout for setup and GitHub publication. External cancellation stops the run at once; unlike the deadline, it does not post completed phases. Empty = 1500
     required: false
     default: "1500"
   max_findings:

diff --git a/fixtures/openrouter.chat-result.json b/fixtures/openrouter.chat-result.json
diff --git a/fixtures/openrouter.invalid-json.json b/fixtures/openrouter.invalid-json.json
diff --git a/fixtures/openrouter.responses-accepted.json b/fixtures/openrouter.responses-accepted.json
@@ -0,0 +1,168 @@
+{
+  "id": "gen-1794440085-n7DSIWy7fBrBUcqMEvla",
+  "object": "response",
+  "created_at": 1794440085,
+  "model": "z-ai/glm-5.3-flash",
+  "status": "completed",
+  "completed_at": 1794440086,
+  "output": [
+    {
+      "id": "rs_tmp_viau8k9rg1",
+      "type": "reasoning",
+      "status": "completed",
+      "content": [
+        {
+          "type": "reasoning_text",
+          "text": "User asks to report no findings. Provide JSON with analysis and empty findings array."
+        }
+      ],
+      "summary": [],
+      "format": "unknown"
+    },
+    {
+      "id": "msg_tmp_o67rrerum0m",
+      "type": "message",
+      "role": "assistant",
+      "status": "completed",
+      "content": [
+        {
+          "type": "output_text",
+          "text": "{\n  \"analysis\": \"Reviewed the single-line change `export const greeting = \\\"hello\\\"`. This is a simple, valid module-level constant export. The syntax is correct, the value is a static string with no side effects, and naming follows common conventions. No security, correctness, style, or maintainability concerns were identified. As requested, there are no findings to report.\",\n  \"findings\": []\n}",
+          "annotations": [],
+          "logprobs": []
+        }
+      ]
+    }
+  ],
+  "error": null,
+  "incomplete_details": null,
+  "access_programs": null,
+  "tools": [],
+  "tool_choice": "auto",
+  "parallel_tool_calls": true,
+  "max_output_tokens": 128000,
+  "temperature": 1,
+  "top_p": 1,
+  "presence_penalty": 0,
+  "frequency_penalty": 0,
+  "top_logprobs": 0,
+  "max_tool_calls": null,
+  "metadata": {},
+  "background": false,
+  "previous_response_id": null,
+  "service_tier": "auto",
+  "truncation": "disabled",
+  "store": false,
+  "instructions": "You are a code reviewer. Reply with JSON that matches the schema: an analysis string and a findings array.",
+  "text": {
+    "format": {
+      "type": "json_schema",
+      "name": "review_response",
+      "strict": true,
+      "schema": {
+        "$schema": "https://json-schema.org/draft/2020-12/schema",
+        "type": "object",
+        "properties": {
+          "analysis": {
+            "type": "string"
+          },
+          "findings": {
+            "type": "array",
+            "items": {
+              "type": "object",
+              "properties": {
+                "file": {
+                  "type": "string",
+                  "minLength": 1
+                },
+                "line": {
+                  "type": "integer",
+                  "exclusiveMinimum": 0,
+                  "maximum": 9007199254740991
+                },
+                "end_line": {
+                  "anyOf": [
+                    {
+                      "type": "integer",
+                      "exclusiveMinimum": 0,
+                      "maximum": 9007199254740991
+                    },
+                    {
+                      "type": "null"
+                    }
+                  ]
+                },
+                "category": {
+                  "type": "string",
+                  "enum": ["correctness", "security", "conventions", "tests", "subtle_bugs", "ci"]
+                },
+                "severity": {
+                  "type": "string",
+                  "enum": ["critical", "high", "medium", "low"]
+                },
+                "confidence": {
+                  "type": "string",
+                  "enum": ["high", "medium", "low"]
+                },
+                "title": {
+                  "type": "string",
+                  "minLength": 1
+                },
+                "description": {
+                  "type": "string",
+                  "minLength": 1
+                },
+                "suggestion": {
+                  "type": ["string", "null"]
+                },
+                "failure_scenario": {
+                  "type": "string",
+                  "minLength": 1
+                }
+              },
+              "required": [
+                "file",
+                "line",
+                "end_line",
+                "category",
+                "severity",
+                "confidence",
+                "title",
+                "description",
+                "suggestion",
+                "failure_scenario"
+              ],
+              "additionalProperties": false
+            }
+          }
+        },
+        "required": ["analysis", "findings"],
+        "additionalProperties": false
+      }
+    }
+  },
+  "reasoning": {
+    "effort": "max"
+  },
+  "safety_identifier": null,
+  "prompt_cache_key": null,
+  "prompt_cache_options": null,
+  "usage": {
+    "input_tokens": 54,
+    "input_tokens_details": {
+      "cached_tokens": 0
+    },
+    "output_tokens": 102,
+    "output_tokens_details": {
+      "reasoning_tokens": 17
+    },
+    "total_tokens": 156,
+    "cost": 0.000021036,
+    "is_byok": false,
+    "cost_details": {
+      "upstream_inference_cost": 0.000021036,
+      "upstream_inference_input_cost": 0.000002268,
+      "upstream_inference_output_cost": 0.000018768
+    }
+  }
+}
diff --git a/fixtures/openrouter.responses-bad-key-body.json b/fixtures/openrouter.responses-bad-key-body.json
@@ -0,0 +1,6 @@
+{
+  "error": {
+    "message": "User not found.",
+    "code": 401
+  }
+}