Merge pull request #15 from foo-software/fix/workflow-run-fallbacks

fix: timestamp fallback in workflow_run mode, PR SHA validation strictness

Author: adamhenson

__tests__/workflow-run.test.ts

@@ -153,7 +153,9 @@ describe('resolveWorkflowRun', () => {
         pull_requests: [],
       },
     });
-    octokit.rest.pulls.list = vi.fn().mockResolvedValue({ data: [{ number: 7 }] });
+    octokit.rest.pulls.list = vi.fn().mockResolvedValue({
+      data: [{ number: 7, head: { sha: '' } }],
+    });
     mockGetOctokit.mockReturnValue(octokit as never);
 
     const result = await resolveWorkflowRun(baseArgs);
@@ -162,6 +164,77 @@ describe('resolveWorkflowRun', () => {
     expect(result.triggerEvent).toBe('pull_request');
   });
 
+  it('selects the SHA-matching PR when multiple PRs share the same branch name', async () => {
+    const octokit = makeOctokit({
+      runData: {
+        run_started_at: '2026-03-09T10:00:00Z',
+        updated_at: '2026-03-09T10:05:00Z',
+        head_branch: 'feat/my-feature',
+        head_sha: 'abc123',
+        event: 'pull_request',
+        pull_requests: [],
+      },
+    });
+    octokit.rest.pulls.list = vi.fn().mockResolvedValue({
+      data: [
+        { number: 10, head: { sha: 'wrong-sha' } },
+        { number: 20, head: { sha: 'abc123' } },
+      ],
+    });
+    mockGetOctokit.mockReturnValue(octokit as never);
+
+    const result = await resolveWorkflowRun(baseArgs);
+
+    expect(result.triggerNumber).toBe(20);
+  });
+
+  it('returns null triggerNumber when headSha is set but no PR head matches', async () => {
+    const octokit = makeOctokit({
+      runData: {
+        run_started_at: '2026-03-09T10:00:00Z',
+        updated_at: '2026-03-09T10:05:00Z',
+        head_branch: 'feat/my-feature',
+        head_sha: 'abc123',
+        event: 'pull_request',
+        pull_requests: [],
+      },
+    });
+    octokit.rest.pulls.list = vi.fn().mockResolvedValue({
+      data: [
+        { number: 10, head: { sha: 'wrong-1' } },
+        { number: 20, head: { sha: 'wrong-2' } },
+      ],
+    });
+    mockGetOctokit.mockReturnValue(octokit as never);
+
+    const result = await resolveWorkflowRun(baseArgs);
+
+    expect(result.triggerNumber).toBeNull();
+  });
+
+  it('falls back to prs[0] when no headSha is available', async () => {
+    const octokit = makeOctokit({
+      runData: {
+        run_started_at: '2026-03-09T10:00:00Z',
+        updated_at: '2026-03-09T10:05:00Z',
+        head_branch: 'feat/my-feature',
+        event: 'pull_request',
+        pull_requests: [],
+      },
+    });
+    octokit.rest.pulls.list = vi.fn().mockResolvedValue({
+      data: [
+        { number: 10, head: { sha: 'sha-1' } },
+        { number: 20, head: { sha: 'sha-2' } },
+      ],
+    });
+    mockGetOctokit.mockReturnValue(octokit as never);
+
+    const result = await resolveWorkflowRun(baseArgs);
+
+    expect(result.triggerNumber).toBe(10);
+  });
+
   it('resolves trigger number from branch name when no pull_requests', async () => {
     const octokit = makeOctokit({
       runData: {

dist/index.js

@@ -99,7 +99,7 @@ When set, `resolveWorkflowRun` does four things:
 
 1. **Gate** — calls `listJobsForWorkflowRun` and exits early unless a job named `conclusion` has completed. Prevents ~5 duplicate ingests from gh-aw's multi-job structure. Single-job workflows pass through unconditionally.
 2. **Status normalization** — maps GitHub conclusions (`failure` → `failed`, `skipped` → skip entirely) to the AgentMeter API enum. Unrecognized statuses (e.g. custom `needs_human`) pass through unchanged.
-3. **Trigger resolution** — reads `pull_requests[]` from the run object; falls back to a `pulls.list` lookup by head branch if empty (GitHub API quirk for some PR-triggered runs). Works for fork PRs. Issue branches are matched only when the branch name contains `agent/issue-N` (the gh-aw convention) — bare `issue-N` patterns are intentionally not matched to avoid misattributing unrelated branches like `feature/fix-issue-12-auth`.
+3. **Trigger resolution** — reads `pull_requests[]` from the run object; falls back to a `pulls.list` lookup by head branch if empty (GitHub API quirk for some PR-triggered runs). When `head_sha` is available, the fallback validates each candidate PR's head SHA against it — if a match is found it is used; if no match is found `triggerNumber` returns `null` rather than guessing. When `head_sha` is absent (API failure), falls back to `prs[0]`. Issue branches are matched only when the branch name contains `agent/issue-N` (the gh-aw convention) to avoid misattributing unrelated branches.
 4. **Token artifact** — downloads and unzips the `agent-tokens` artifact using `fflate`.
 
 ### Pricing (`src/pricing.ts`)
@@ -125,7 +125,7 @@ When `workflow_run_id` is provided, `githubRunId` in the ingest payload is set t
 
 `context.ts` maps GitHub event names to AgentMeter trigger types. `issue_comment` is correctly classified as `pr_comment` when `payload.issue.pull_request` is present, and `issue_comment` otherwise.
 
-In companion `workflow_run` mode, `resolveTrigger` in `workflow-run.ts` returns a `triggerType` and `triggerRef` that reflect the original triggering event, not the companion workflow's own `workflow_run` event.
+In companion `workflow_run` mode, timestamps resolve from the agent run API (`run_started_at`, `updated_at`). If the API call fails, timestamps are left empty and `durationSeconds` resolves to 0 — the companion workflow's own start time is never substituted to avoid recording a misleading duration.
 
 ---
 

docs/agentmeter-action-spec.md

@@ -82,9 +82,18 @@ export async function run(): Promise<void> {
       // Use normalized status from the run data
       inputs = { ...inputs, status: runData.normalizedStatus };
 
-      // Only override with resolved values when explicit inputs aren't set
-      if (!inputs.startedAt && runData.startedAt) resolvedStartedAt = runData.startedAt;
-      if (!inputs.completedAt && runData.completedAt) resolvedCompletedAt = runData.completedAt;
+      // Only override with resolved values when explicit inputs aren't set.
+      // In workflow_run mode never fall back to selfStartedAt/now — those are the companion
+      // workflow's times, not the agent run's times. Use empty string so durationSeconds
+      // safely resolves to 0 rather than silently recording the wrong run's duration.
+      resolvedStartedAt = inputs.startedAt || runData.startedAt || '';
+      resolvedCompletedAt = inputs.completedAt || runData.completedAt || '';
+      if (
+        (!inputs.startedAt && !runData.startedAt) ||
+        (!inputs.completedAt && !runData.completedAt)
+      ) {
+        core.warning('AgentMeter: workflow run timestamps unavailable — duration will be omitted.');
+      }
       if (inputs.triggerNumber === null) resolvedTriggerNumber = runData.triggerNumber;
       if (!inputs.triggerEvent) resolvedTriggerEvent = runData.triggerEvent;
       resolvedTriggerRef = runData.triggerRef;

src/run.ts

@@ -314,7 +314,9 @@ async function resolveTrigger({
       });
       // Validate by head SHA when available to avoid matching the wrong PR when multiple
       // PRs share the same branch name (e.g. reused or fork branches).
-      const match = headSha ? (prs.find((pr) => pr.head.sha === headSha) ?? prs[0]) : prs[0];
+      // When headSha is provided but no PR matches, return null rather than guessing.
+      const shaMatch = headSha ? prs.find((pr) => pr.head.sha === headSha) : null;
+      const match = shaMatch ?? (headSha ? null : prs[0]);
       if (match) {
         return {
           triggerNumber: match.number,