fix(results): harden static report publishing (#1373)

Author: christso

apps/cli/src/commands/results/report.ts

@@ -124,7 +124,7 @@ export function renderResultsReport(
     serializeReportResult(result, sourceFile, records[index], benchmarkEvalFile),
   );
   const dataJson = JSON.stringify(rows).replace(/<\//g, '<\\/');
-  return RESULTS_REPORT_TEMPLATE.replace('__DATA_PLACEHOLDER__', dataJson);
+  return RESULTS_REPORT_TEMPLATE.replace('__DATA_PLACEHOLDER__', () => dataJson);
 }
 
 export async function writeResultsReport(

apps/cli/test/commands/results/report.test.ts

@@ -148,6 +148,27 @@ describe('results report', () => {
     expect(html).not.toContain('Grader Results');
   });
 
+  it('embeds result text containing replacement tokens without corrupting the inline script', async () => {
+    const runDir = path.join(tempDir, 'run');
+    await writeArtifactsFromResults(
+      [
+        makeResult({
+          output: "literal replacement tokens: $' $& $` $$",
+        }),
+      ],
+      runDir,
+      { evalFile: 'evals/demo.eval.yaml' },
+    );
+
+    const { outputPath } = await writeResultsReport(runDir, undefined, tempDir);
+    const html = readFileSync(outputPath, 'utf8');
+
+    expect(html.match(/<\/script>/g)).toHaveLength(1);
+    expect(html).toContain("literal replacement tokens: $' $& $` $$");
+    expect(html).toContain('<script>');
+    expect(html).toContain('</script>');
+  });
+
   it('emits an inline report script that parses successfully', async () => {
     const runDir = path.join(tempDir, 'run');
     await writeArtifactsFromResults([makeResult()], runDir, { evalFile: 'evals/demo.eval.yaml' });

apps/web/src/content/docs/docs/tools/results.mdx

@@ -54,6 +54,31 @@ What it shows:
 - **Expandable details** — unified assertions with pass/fail indicators and type badges, collapsible input/output
 - **Criteria column** — shows the test prompt or description inline for quick scanning
 
+### Publish a static report with GitHub Pages
+
+The generated file is self-contained HTML: no Dashboard server, API endpoint, or external asset host is required after it is written. That makes it a good fit for public result repositories served by GitHub Pages.
+
+One minimal publication workflow is:
+
+```bash
+# 1. Run an eval and sync or copy the run workspace into your public results repo.
+agentv eval evals/demo.eval.yaml --output .agentv/results/runs/demo-live
+
+# 2. In the public results repo, render the report into the Pages source directory.
+agentv results report .agentv/results/runs/demo-live --out docs/index.html
+
+# 3. Review the generated HTML before publishing.
+grep -RInE 'sk-[A-Za-z0-9]|Bearer |localhost|127\.0\.0\.1|/home/|/Users/|/tmp/' docs/index.html
+
+# 4. Commit the run artifacts and docs/index.html, then enable GitHub Pages
+#    for the repository's docs/ directory or the branch used for Pages.
+git add .agentv/results/runs/demo-live docs/index.html README.md
+git commit -m "docs(results): publish static AgentV report"
+git push
+```
+
+Use `--out docs/<name>.html` when a repository should publish multiple runs. Link those files from the result repository README so readers can browse a dashboard-like report from GitHub Pages instead of running `agentv dashboard` or opening raw JSONL.
+
 <Image src={resultsReportDetails} alt="AgentV results report showing an expanded failing test case with unified assertions, deterministic type badges, pass/fail indicators, evidence text, and collapsible input/output" />
 
 | Option | Description |

docs/brainstorms/2026-06-14-static-results-dashboard-recommendation.md

@@ -0,0 +1,125 @@
+---
+date: 2026-06-14
+topic: static-results-dashboard
+type: recommendation
+---
+
+# Static Results Dashboard Recommendation
+
+## Summary
+
+AgentV should keep `agentv results report` focused on one run and add any future multi-run GitHub Pages export as a separate `agentv results dashboard` command. The dashboard should be a read-only static index over existing run artifacts, with relative links to single-run reports for drill-down.
+
+---
+
+## Problem Frame
+
+Public benchmark repositories need a URL that explains quality movement over time without asking readers to clone artifacts, run `agentv dashboard`, or interpret raw `index.jsonl`. The existing single-run report solves the case-level review problem, but it does not answer longitudinal questions such as whether pass rate drifted, which target regressed, or whether a release improved a benchmark pack.
+
+AgentV already has the core primitives for those questions: canonical run workspaces, `agentv trend` for score drift, `agentv compare` for pairwise and target matrix deltas, and Dashboard analytics for run lists, score distribution, normalized gain, tag heatmaps, and trend charts. The static dashboard should compose those existing primitives into a publishable artifact rather than creating a second hosted Dashboard.
+
+---
+
+## Key Decisions
+
+- **New command, not `results report` mode expansion.** `results report` has a clear one-run contract and one default output file. A multi-run dashboard has different source semantics, writes an index plus optional per-run pages, and should not add a second mode to the report command.
+- **Run artifacts remain canonical.** The static dashboard reads existing `.agentv/results/runs/**/index.jsonl` and summary artifacts from public result repos. It does not introduce a committed database, service worker, or new canonical index.
+- **Static-first output.** The primary artifact is `docs/index.html` with inline data and no backend. Optional drill-down pages use safe relative links, such as `docs/runs/<run-slug>.html`.
+- **Graceful sparse data.** Legal currently has one published live run, while financial has multiple useful runs. Trend and comparison panels should render when data supports them and otherwise show an explanatory empty state.
+
+---
+
+## Requirements
+
+**Inputs**
+
+- R1. The dashboard command accepts one or more result repo roots, run workspace directories, or `index.jsonl` manifests.
+- R2. The command discovers canonical run workspaces under `.agentv/results/runs/` when given a repo root.
+- R3. The command uses existing run metadata and `index.jsonl` rows without requiring a live Dashboard or remote API.
+- R4. Optional filtering supports suite, target, experiment, tags, and latest-N run selection when those fields exist in artifacts.
+
+**Output**
+
+- R5. The default output is a GitHub Pages-friendly `docs/index.html` that is read-only and self-contained.
+- R6. Per-run drill-down links are relative and can point to generated single-run report pages.
+- R7. The generated HTML uses the existing Dashboard/report visual language: dark surface hierarchy, summary cards, pass/fail tones, score distribution, run table, and expandable details through linked reports.
+- R8. The output includes no external scripts, no external stylesheets, no local endpoint references, and no local filesystem paths.
+
+**Analysis**
+
+- R9. The dashboard summarizes total runs, latest run, pass rate, mean score, failures, errors, duration, token usage, and cost when present.
+- R10. The dashboard shows pass-rate and mean-score trend lines across runs when at least two comparable runs exist.
+- R11. The dashboard shows score distributions for the selected run set.
+- R12. The dashboard shows target/model comparison tables when multiple targets are present.
+- R13. The dashboard surfaces release/regression signals by highlighting latest-run deltas against a selected baseline or previous run.
+
+**Publication**
+
+- R14. The command supports a public results repository workflow where `docs/index.html` is committed with run artifacts and served by GitHub Pages.
+- R15. The command can optionally generate single-run report pages for every included run by reusing the existing report renderer.
+- R16. The command emits enough terminal output for automation to record the generated files and source run count.
+
+---
+
+## MVP for Legal and Financial Demos
+
+The first useful version does not need a full Dashboard clone. It needs:
+
+1. `agentv results dashboard <results-repo-root> --out docs/index.html --reports docs/runs/`
+2. A static overview with cards for runs, tests, latest pass rate, mean score, failures, and errors.
+3. A run table with date, experiment/run ID, target, tests, pass rate, mean score, and a relative `runs/<slug>.html` report link.
+4. A score distribution histogram over the selected runs.
+5. A pass-rate trend chart when at least two comparable runs exist.
+6. A target comparison table when at least two targets exist.
+7. Empty states that explain why legal has no trend yet if only one legal run is published.
+
+For the current legal and financial publication task, the existing `agentv results report` primitive is enough: publish `docs/index.html` in the legal results repo, and publish `docs/index.html` plus `docs/dexter-baseline.html` in the financial results repo. The multi-run dashboard should be a follow-up because it would add a new command, source discovery rules, and aggregation behavior beyond this issue's smallest GitHub Pages-ready path.
+
+---
+
+## URL and Layout Shape
+
+```text
+docs/
+  index.html                       # multi-run static dashboard
+  runs/
+    <experiment>-<timestamp>.html   # generated with the single-run report renderer
+```
+
+Suggested tabs or sections for `docs/index.html`:
+
+- **Overview** — cards, latest run, selected filters, and publication timestamp.
+- **Trends** — pass-rate and mean-score lines, with sparse-data empty states.
+- **Compare** — target/model matrix and latest-vs-baseline deltas.
+- **Runs** — sortable run table with relative links to generated single-run reports.
+
+---
+
+## Scope Boundaries
+
+- Do not build a second live Dashboard app.
+- Do not require a server, database, API route, or GitHub Actions service to view the output.
+- Do not commit a generated SQLite database or append-only aggregate index as canonical state.
+- Do not replace `agentv trend`, `agentv compare`, or Dashboard analytics; reuse their concepts and data shapes.
+- Do not add statistical claims beyond descriptive trend/delta summaries in the MVP.
+
+---
+
+## Acceptance Examples
+
+- AE1. **Covers R1, R2, R5.** Given a public results repo root, when the command runs with `--out docs/index.html`, then a static dashboard file is generated from discovered run workspaces.
+- AE2. **Covers R6, R15.** Given `--reports docs/runs/`, when the command includes a run, then the run table links to a relative single-run report page.
+- AE3. **Covers R10.** Given fewer than two comparable legal runs, when the dashboard renders, then the trend panel explains that more runs are required instead of showing a misleading chart.
+- AE4. **Covers R8, R14.** Given the generated `docs/` directory is scanned, then no external scripts, local endpoints, local filesystem paths, or secrets are present.
+- AE5. **Covers R12, R13.** Given financial runs include multiple targets or a baseline/candidate pair, when the dashboard renders, then target comparison or latest-vs-baseline deltas are visible without running the live Dashboard.
+
+---
+
+## Sources
+
+- `apps/cli/src/commands/results/report.ts` — existing one-run static HTML report writer.
+- `apps/cli/src/commands/results/report-template.ts` — existing Dashboard-themed static report visual language.
+- `apps/cli/src/commands/trend/index.ts` — existing multi-run trend analysis over run manifests.
+- `apps/cli/src/commands/compare/index.ts` — existing pairwise and target-matrix comparison logic.
+- `apps/web/src/content/docs/docs/tools/dashboard.mdx` — current Dashboard analytics behavior to mirror at static-export scope.
+- `apps/web/src/content/docs/docs/tools/results.mdx` — current single-run report docs and GitHub Pages workflow.