Problem

When collaborating via crit-web, the local review session doesn't reflect changes after the AI agent acts on web feedback. The current flow requires manual intervention:

1. crit <args> → local review
2. crit share → publish to web
3. Reviewer comments on crit-web, clicks "Send to author"
4. Human pastes prompt into local AI
5. AI makes code changes + crit share
6. Human's local crit session is stale — doesn't show new code state

Solution

Three changes to close the loop automatically:

1. Add cli_args field to CritJSON

Store the original CLI args in the review session so they can round-trip through crit-web.

2. Include cli_args in crit share payload

When sharing, send the original args alongside the review data.

3. Update crit-web prompt generation

When generating the "Send to author" prompt, append crit <cli_args> after crit share. The prompt changes from:

make changes and respond, then crit share

to:

make changes, then crit share, then crit <original args>

Result

The human pastes one prompt, the AI makes changes, shares to web, and re-runs crit <original args> which refreshes the local browser session with the new code state. No manual steps needed.

Touches

• crit/ — CritJSON struct + share payload
• crit-web/ — prompt generation template