openclaw/docs/tools/trajectory.md

summary, read_when, title

summary	read_when	title
Export redacted trajectory bundles for debugging an OpenClaw agent session	Debugging why an agent answered, failed, or called tools a certain way Exporting a support bundle for an OpenClaw session Investigating prompt context, tool calls, runtime errors, or usage metadata Disabling or relocating trajectory capture	Trajectory bundles

Trajectory capture is OpenClaw's per-session flight recorder. It records a structured timeline for each agent run, then /export-trajectory packages the current session into a redacted support bundle.

Use it when you need to answer questions like:

• What prompt, system prompt, and tools were sent to the model?
• Which transcript messages and tool calls led to this answer?
• Did the run time out, abort, compact, or hit a provider error?
• Which model, plugins, skills, and runtime settings were active?
• What usage and prompt-cache metadata did the provider return?

If you are filing a broad support report for a live Gateway issue, start with /diagnostics. Diagnostics collects the sanitized Gateway bundle and, for OpenAI Codex harness sessions, can also send Codex feedback to OpenAI servers after approval. Use /export-trajectory when you specifically need the detailed per-session prompt, tool, and transcript timeline.

Quick start

Send this in the active session:

/export-trajectory

Alias:

/trajectory

OpenClaw writes the bundle under the workspace:

.openclaw/trajectory-exports/openclaw-trajectory-<session>-<timestamp>/

You can choose a relative output directory name:

/export-trajectory bug-1234

The custom path is resolved inside .openclaw/trajectory-exports/. Absolute paths and ~ paths are rejected.

Trajectory bundles can contain prompts, model messages, tool schemas, tool results, runtime events, and local paths. The chat slash command therefore runs through exec approval every time. Approve the export once when you intend to create the bundle; do not use allow-all. In group chats, OpenClaw sends the approval prompt and export result to the owner privately instead of posting the trajectory details back to the shared room.

For local inspection or support workflows, you can also run the approved command path directly:

openclaw sessions export-trajectory --session-key "agent:main:telegram:direct:123" --workspace .

Access

Trajectory export is an owner command. The sender must pass the normal command authorization checks and owner checks for the channel.

What gets recorded

Trajectory capture is on by default for OpenClaw agent runs.

Runtime events include:

• session.started
• trace.metadata
• context.compiled
• prompt.submitted
• model.fallback_step, including the source model, next model, failure reason/detail, chain position, and whether fallback advanced, succeeded, or exhausted the chain
• model.completed
• trace.artifacts
• session.ended

Transcript events are also reconstructed from the active session branch:

• user messages
• assistant messages
• tool calls
• tool results
• compactions
• model changes
• labels and custom session entries

Events are written as JSON Lines with this schema marker:

{
  "traceSchema": "openclaw-trajectory",
  "schemaVersion": 1
}

Bundle files

An exported bundle can contain:

File	Contents
manifest.json	Bundle schema, source files, event counts, and generated file list
events.jsonl	Exported ordered runtime and transcript timeline
session-branch.json	Redacted active transcript branch and session header
metadata.json	OpenClaw version, OS/runtime, model, config snapshot, plugins, skills, and prompt metadata
artifacts.json	Final status, errors, usage, prompt cache, compaction count, assistant text, tool metadata, and SQLite tool-artifact metadata
prompts.json	Submitted prompts and selected prompt-building details
system-prompt.txt	Latest compiled system prompt, when captured
tools.json	Tool definitions sent to the model, when captured

manifest.json lists the files present in that bundle. Some files are omitted when the session did not capture the corresponding runtime data.

artifacts.json may include toolArtifacts entries for run-scoped SQLite artifacts such as runtime trajectory mirrors or tool media manifests. These entries are metadata-only: the export omits artifact blobs and blobBase64 payloads so large generated media is not duplicated into the support bundle.

Capture location

By default, runtime trajectory events are written to the owning agent database:

~/.openclaw/agents/<agentId>/agent/openclaw-agent.sqlite
trajectory_runtime_events

The export manifest reports this source as structured database provenance, for example:

{
  "sourceDatabases": {
    "runtime": {
      "role": "agent",
      "agentId": "<agentId>",
      "table": "trajectory_runtime_events",
      "sessionId": "<sessionId>"
    }
  }
}

/export-trajectory reads runtime events from SQLite and materializes events.jsonl only inside the explicit support bundle. New runtime captures do not create legacy <session>.trajectory.jsonl sidecars or pointer files.

Disable capture

Set OPENCLAW_TRAJECTORY=0 before starting OpenClaw:

export OPENCLAW_TRAJECTORY=0

This disables runtime trajectory capture. /export-trajectory can still export the transcript branch, but runtime-only files such as compiled context, provider artifacts, and prompt metadata may be missing.

Privacy and limits

Trajectory bundles are designed for support and debugging, not public posting. OpenClaw redacts sensitive values before writing export files:

• credentials and known secret-like payload fields
• image data
• local state paths
• workspace paths, replaced with $WORKSPACE_DIR
• home directory paths, where detected

The exporter also bounds input size:

• runtime capture: live capture stops at 10 MiB and records a truncation event when space remains
• transcript branch export: 50 MiB
• runtime events: 200,000
• total exported events: 250,000
• individual runtime event lines are truncated above 256 KiB

Review bundles before sharing them outside your team. Redaction is best-effort and cannot know every application-specific secret.

Troubleshooting

If the export has no runtime events:

• confirm OpenClaw was started without OPENCLAW_TRAJECTORY=0
• run another message in the session, then export again
• inspect manifest.json for runtimeEventCount

If the command rejects the output path:

• use a relative name like bug-1234
• do not pass /tmp/... or ~/...
• keep the export inside .openclaw/trajectory-exports/

If the export fails with a size error, the transcript branch or runtime capture exceeded the export safety limits. Start a new session or export a smaller reproduction.

Related

• Diffs
• Session management
• Exec tool