Skip to content

DASHBOARD.md

Latest commit

 

History

History
136 lines (104 loc) · 4.79 KB

DASHBOARD.md

File metadata and controls

136 lines (104 loc) · 4.79 KB
Raw

Dashboard

AgentOps Workbench includes a local dashboard foundation backed by the same SQLite store used by the CLI.

The dashboard is supported as a local UI. Version v1.2.0 adds decision views for merge readiness, claim/evidence checks, risk drilldown, evidence exports, and run comparison. The v1.4 forensic import work adds evidence quality labels for plain-text transcript imports. Browser JSON endpoints remain local implementation details, not a stable remote API. See Compatibility policy.

Start

Ingest synthetic demo sessions:

./bin/agentops audit ./fixtures/sample-session.jsonl
./bin/agentops audit ./fixtures/needs-review-session.jsonl
./bin/agentops audit ./fixtures/risky-session.jsonl
./bin/agentops audit ./fixtures/forensic-terminal-transcript.txt

Start the local server:

./bin/agentops open

Default URL:

http://127.0.0.1:4927

Use a different local port when needed:

./bin/agentops open --port 4930

Validate the dashboard configuration without starting a long-running server:

./bin/agentops open --check

Scope

The dashboard currently provides:

  • session list
  • session text and adapter filtering
  • evidence-quality summary for structured JSONL and forensic text imports
  • merge-readiness summary for a selected session
  • claim-vs-evidence matrix for tests, lint, typecheck, build, and final success, with separate labels for verified, inferred, and missing evidence
  • risk severity drilldown with linked event, command, file, and evidence context
  • run comparison between two sessions
  • session timeline
  • command drilldown
  • file-change drilldown
  • MCP and tool usage summary
  • token summary when available
  • Markdown report export for the selected session
  • sanitized JSON evidence export for the selected session

It reads from the local SQLite database only. It does not require hosted services, does not post to GitHub, and does not upload session data. Dashboard evidence exports omit source artifact paths, command output, raw event JSON, and raw payload hashes to reduce local-environment leakage in demos and screenshots.

Demo States

The public fixtures are synthetic and cover the dashboard's main decision states:

State Fixture Expected Dashboard Signal
Empty no ingested sessions empty session list and no selected session
Ready fixtures/sample-session.jsonl no risks and bun test evidence
Needs review fixtures/needs-review-session.jsonl final success claim without verification evidence
Blocked fixtures/risky-session.jsonl high risks, medium risks, and missing success evidence
Forensic fixtures/forensic-terminal-transcript.txt forensic-text adapter, observed shell commands, inferred file changes
Inferred forensic fixtures/forensic-copied-chat.txt inferred command evidence requires review instead of showing as fully verified
Comparison sample-session against risky-session reduced risks and added verification

API

The local server exposes JSON endpoints used by the browser UI:

GET /api/health
GET /api/sessions?limit=50
GET /api/sessions/:id
GET /api/sessions/:id/report
GET /api/sessions/:id/evidence
GET /api/compare?base=<session-id>&target=<session-id>

The report endpoint returns the same Markdown session report generated by agentops save report <id>. The evidence endpoint returns a sanitized JSON bundle for local review artifacts. The comparison endpoint compares two stored sessions and is used by the dashboard's run comparison panel.

These endpoints are intended for local dashboard use. They are not a stable remote API contract yet.

Privacy

Dashboard demos and screenshots should use synthetic fixtures only.

Do not use real private transcripts, private PAI memory, local personal paths, credentials, or sensitive command output in public screenshots.

Browser Verification

Basic manual verification:

  1. Run ./bin/agentops audit ./fixtures/sample-session.jsonl.
  2. Run ./bin/agentops audit ./fixtures/needs-review-session.jsonl.
  3. Run ./bin/agentops audit ./fixtures/risky-session.jsonl.
  4. Run ./bin/agentops open.
  5. Open http://127.0.0.1:4927.
  6. Confirm the session list shows sample-session, needs-review-session, and risky-session.
  7. Confirm evidence quality, merge readiness, claim/evidence, risk drilldown, timeline, commands, files, tools, and token metrics render.
  8. Select a comparison session from the Compare with control and confirm the run comparison panel renders.
  9. Open the Markdown report and JSON evidence links for the selected session.

Automated coverage verifies the dashboard JSON endpoints, representative demo states, comparison payload, evidence export hygiene, and HTML shell. Visual screenshots should be captured from synthetic fixtures only.