Skip to content

Count completed Agent jobs in the daily AI Credits guardrail #66302

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pelikhan merged 6 commits into from
Oct 7, 2026
Merged

Count completed Agent jobs in the daily AI Credits guardrail #66302

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
db87e18
Initial plan
Copilot Oct 6, 2026
7e79ef5
Count displayed AIC component jobs and refresh coverage cache
Copilot Oct 6, 2026
833f1fe
Merge remote-tracking branch 'origin/main' into copilot/fix-daily-aic…
Copilot Oct 7, 2026
86450fe
Count Evaluations job usage in daily AIC guardrail
Copilot Oct 7, 2026
520e8c9
Merge remote-tracking branch 'origin/main' into copilot/fix-daily-aic…
Copilot Oct 7, 2026
2c392c1
Verify daily AIC job identifiers take precedence over names
Copilot Oct 7, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions actions/setup/js/daily_aic_cache_helpers.cjs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ function readScanCache(content, repository, workflowId, now = Date.now()) {
}
if (
entry?.version !== 2 ||
entry.coverage_version !== 1 ||
entry.coverage_version !== 2 ||
entry.repository !== repository ||
entry.workflow_id !== workflowId ||
!Number.isSafeInteger(entry.run_id) ||
Expand Down Expand Up @@ -72,7 +72,7 @@ function scanCacheEntry(run, aic, repository, workflowId, now = Date.now(), sour
}
return {
version: 2,
coverage_version: 1,
coverage_version: 2,
repository,
workflow_id: workflowId,
run_id: run.id,
Expand Down
7 changes: 4 additions & 3 deletions actions/setup/js/daily_aic_component_coverage.cjs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,16 +28,17 @@ async function loadBillableJobs({ github, budget }, owner, repo, run) {
const jobs = response.data.jobs;
if (!Array.isArray(jobs)) throw new Error("Incomplete daily AIC job metadata");
for (const job of jobs) {
if (!Object.hasOwn(COMPONENT_FILES, job.name)) continue;
const name = typeof job.name === "string" ? job.name.toLowerCase() : "";

@github-actions github-actions Bot Oct 6, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This fix correctly lowercases GitHub's display names Agent/Detection to match COMPONENT_FILES, but the real compiled evals job display name is Evaluations, not Evals (confirmed across all 167 compiled .github/workflows/*.lock.yml — e.g. ab-testing-advisor.lock.yml line 1913: name: Evaluations).

"evaluations".toLowerCase() does not match the evals key in COMPONENT_FILES (actions/setup/js/daily_aic_component_coverage.cjs:9-12), so completed Evaluations jobs will still be silently skipped by Object.hasOwn(COMPONENT_FILES, name) — the exact same class of bug this PR is fixing for Agent/Detection, just left unfixed for evals.

The new regression test at daily_aic_component_coverage.test.cjs:108 uses job("Evals") instead of the real display name job("Evaluations"), which masks this gap (the fabricated name happens to lowercase to evals and pass).

Suggest adding an evaluations → evals alias (or mapping GitHub's actual display names explicitly) and updating the test to use the real compiled name "Evaluations".

@copilot please address this.

Copilot AI Oct 7, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Handled in 86450fe: Evaluations now maps to evals for component and artifact accounting, with regression coverage using the actual display name.

if (!Object.hasOwn(COMPONENT_FILES, name)) continue;
if (!Number.isSafeInteger(job.run_attempt) || job.run_attempt < 1 || job.run_attempt > run.run_attempt || job.status !== "completed" || !job.conclusion) {
throw new Error("Incomplete daily AIC component attempt metadata");
}
const prior = components.get(job.name);
const prior = components.get(name);
if (prior && job.run_attempt === prior.run_attempt && job.id !== prior.id) {
throw new Error("Ambiguous daily AIC component jobs");
}
if (!prior || (prior.conclusion === "skipped" && job.conclusion !== "skipped") || (job.conclusion !== "skipped" && job.run_attempt > prior.run_attempt) || (prior.conclusion === "skipped" && job.run_attempt > prior.run_attempt)) {
components.set(job.name, job);
components.set(name, job);
}
}
if (jobs.length < 100) {
Expand Down
19 changes: 19 additions & 0 deletions actions/setup/js/daily_aic_component_coverage.test.cjs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,6 +105,25 @@ it("counts an empty detection accounting file as zero AIC", async () => {
expect(global.core.info).toHaveBeenCalledWith(expect.stringContaining('"source":"detection/token_usage.jsonl"'));
});

it("counts compiler-generated Agent, Detection and Evals jobs using their lowercase accounting paths", async () => {
const f = evaluate(
{
"agent/token_usage.jsonl": '{"aic":18.02604}',
"detection/token_usage.jsonl": '{"aic":30.08748}',
"evals/token_usage.jsonl": '{"aic":1}',
},
[job("Agent"), job("Detection"), job("Evals")]

@github-actions github-actions Bot Oct 6, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[/tdd] The compiler's real generated evals job name is "Evaluations" (see pkg/workflow/compiler_job_annotations.go:17, confirmed as name: Evaluations in every compiled .lock.yml), not "Evals". "Evaluations".toLowerCase() is "evaluations", which does not match the evals key in COMPONENT_FILES, so this fix does not actually resolve evals-job accounting — only agent/detection are genuinely fixed.

💡 Why this matters

The test at line 115 (job("Evals")) exercises a job name that never occurs in a real workflow run, so it gives false confidence that evals coverage is fixed. In production, evals jobs will still be silently skipped (zero-credit) exactly like before this PR, defeating the PR's stated goal of fixing counted jobs across the board.

Suggested fix: either special-case "evaluations" in the lookup (e.g. add an alias map from display name → component key), or derive the lowercase key more robustly, and add a regression test using the actual display name "Evaluations".

@copilot please address this.

Copilot AI Oct 7, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Handled in 86450fe: the test now uses the compiler-generated Evaluations name, which is normalized to the lowercase evals accounting component.

);
await expect(f.result).resolves.toBeCloseTo(49.11352);
expect(f.client.listArtifacts).toHaveBeenCalledOnce();
expect(global.core.info).not.toHaveBeenCalledWith(expect.stringContaining('"reason":"no_billable_jobs"'));
});

it("treats differently cased names for the same component as ambiguous", async () => {
const f = evaluate({}, [job("Agent", { id: 1 }), job("agent", { id: 2 })]);
await expect(f.result).rejects.toThrow("Ambiguous daily AIC component jobs");
});

it("still requires accounting when detection/token_usage.jsonl is missing (not empty)", async () => {
const f = evaluate(
{
Expand Down
35 changes: 35 additions & 0 deletions actions/setup/js/daily_aic_scan.test.cjs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,6 +123,41 @@ describe("complete daily AIC scan observations", () => {
expect(global.core.info).toHaveBeenCalledWith(expect.stringContaining('"reason":"scan_cache"'));
});

it("recounts version 1 zero-credit observations from completed Agent and Detection jobs", async () => {
const prior = scanCacheEntry(run(1), 0, repository, 7, now);
prior.coverage_version = 1;
writeEntries([prior]);
const f = fixture([run(1)]);
f.getRunAIC = guardrail.getRunAIC;
f.github.rest.actions.listJobsForWorkflowRun.mockResolvedValue(
response({
jobs: [
{ id: 1, name: "Agent", run_attempt: 1, status: "completed", conclusion: "success", started_at: time, completed_at: time },
{ id: 2, name: "Detection", run_attempt: 1, status: "completed", conclusion: "success", started_at: time, completed_at: time },
],
})
);
f.artifactClient = {
listArtifacts: vi.fn(async () => ({ artifacts: [{ id: 10, name: "usage", createdAt: new Date(time) }] })),
downloadArtifact: vi.fn(async (_id, options) => {
for (const [name, aic] of [
["agent", 18.02604],
["detection", 30.08748],
]) {
const file = path.join(options.path, name, "token_usage.jsonl");
fs.mkdirSync(path.dirname(file), { recursive: true });
fs.writeFileSync(file, JSON.stringify({ aic }));
}
return { downloadPath: options.path };
}),
};
const result = await scanDailyAIC(f);
expect(result.cacheHits).toBe(0);
expect(result.countedRuns[0].aic).toBeCloseTo(48.11352);
expect(readScanCache(fs.readFileSync(cachePath, "utf8"), repository, 7, now).get(1)).toMatchObject({ coverage_version: 2, source: "recorded" });
expect((await scanDailyAIC(f)).cacheHits).toBe(1);
});

it.each(["missing", "metadata-only", "malformed", "unknown-model", "old-attempt", "detection", "evals", "invalid-numeric"])("conservatively assumes max AI Credits when %s usage cannot be resolved", async kind => {
const f = fixture([run(1)]);
f.getRunAIC = guardrail.getRunAIC;
Expand Down
Loading