The walkthrough uses a small task-manager project to compare tools that make coding-agent activity easier to inspect. Filetree exposes project navigation, while Replay Theater displays recent edit requests. The narrator distinguishes those displays from evidence that an edit actually reached disk, checking the resulting diff and tests instead.
Blast Radius adds a checkpoint for recognized risky shell commands, and Collision Guard warns about overlapping edits from participating local sessions. Neither is presented as complete protection: scripts, other tools and untracked changes can escape their coverage. AgentFlow adds visibility into delegated agents but does not make their model work free or automatically faster.
Reflect turns repeated corrections into instructions that a person reviews before saving at the appropriate scope. The cache-warming example receives the most qualified treatment: cache duration varies by provider, warming itself consumes tokens, mocked tests do not prove live cache retention, and estimated API-equivalent costs are not a direct subscription-quota measurement. The practical recommendation is to test the extension that addresses an actual problem rather than enable all seven.
Watch on YouTube
