writ review

Review the latest run in the current project:

writ review

This is the normal review command after writ connect, writ start, or writ issue.

For an attached run that used a tool, writ finish --agent <agent> adds a human satisfaction attestation. If no tool was used, it cancels the waiting activation without an attestation. writ finish --agent <agent> --cancel adds an interruption and does not attest satisfaction.

Select a run

Name an agent to review its latest connected session:

writ review codex

Use claude, codex, cursor, or opencode.

You can also name a writ or an exact run ID:

writ review bug-fixer
writ review <run-id>

Without a target, Writ selects the run with the most recent recorded activity.

Filter the transcript

Use --where more than once to combine filters with AND.

Form	Example
type=<entry type>	--where type=tool_call
actor=<actor>	--where actor=judge
status=<status>	--where status=blocked
text~<text>	--where text~pytest
seq>=<number>	--where seq>=40
seq<=<number>	--where seq<=80

writ review codex \
  --where actor=holder \
  --where status=blocked

The header meters cover the complete run. Filters change only the transcript lines.

Understand argument storage

A normal tool-call entry contains normalized arguments. It contains a digest of exact provider arguments, not the exact provider arguments. Exact provider arguments stay transient unless Writ creates an explicit approval request. The approval request can store those arguments after redaction.

Review can show the redacted arguments of an approval request. Redaction reduces accidental disclosure. It does not make every value safe to share.

Use the local browser view

writ review --web

The default address is http://localhost:8321. The page refreshes while the run is active. It is read-only. It loads no scripts or external assets.

Use --addr to select another listen address:

writ review codex --web --addr localhost:9000

Use writ record to export, witness, or independently verify a signed record.