Agent Flight Recorder

Agents make changes quickly, but the path they took is hard to inspect after the fact.

The Flight Recorder wraps your agent command and records the full timeline — prompt, model, every tool call, every shell command, errors, retries, the final diff, and an estimated cost.

Inspect a run, replay context, export it as HTML, or diff two runs against each other.

• What prompt started this run?
• Which agent and model were used?
• What files did the agent read or edit?
• What commands did it run? Did tests fail before they passed?
• Did the agent retry or loop?
• How long did the run take and what did it cost?

One command. No account. No telemetry.

1# install (binary is `agent-flight`)
2npm install -g @agentopssec/agent-flight-recorder
3 
4# wrap any agent command
5agent-flight run -- codex "fix the failing tests"
6agent-flight run -- gemini "refactor this component"
7 
8# inspect later
9agent-flight inspect latest
10agent-flight export run_001 --html

A real run from the test suite — fixing failing tests in a Next.js app.

Stored locally as JSON. Used by Agent Review and Agent Cost Lens.

1{ "runId": "run_001",
2  "agent": "codex", "prompt": "fix the failing tests",
3  "events": [
4    { "type": "shell.exec", "command": "npm test", "exitCode": 1, "durationMs": 12944 },
5    { "type": "file.change", "path": "src/lib/parser.ts", "changeType": "modified" }
6  ],
7  "gitDiffSummary": { "filesChanged": 3, "insertions": 44, "deletions": 18 },
8  "estimatedCost": 0.18 }