Skip to content
Kapacitor docs

Errors & plan validation

Two adjacent post-session commands: kapacitor errors finds where things went wrong, kapacitor validate-plan confirms the agent finished what it planned. Both work on any session, the active one or a past one.

kapacitor errors

Section titled “kapacitor errors”

Scan a session for tool-call errors — failed bash commands, file read/write errors, agent failures.

Terminal window
kapacitor errors <sessionId>              # one session
kapacitor errors <meta-session-slug>      # meta session (full chain by slug)
kapacitor errors --chain <sessionId>      # full continuation chain

Output is grouped by tool with a short description of each failure. Useful for:

  • Post-session review: spot recurring mistakes and update project instructions.
  • Debugging a stuck agent: see whether it kept retrying the same failing operation.
  • Auditing autonomy: which destructive commands the agent ran, and which were rejected.

With the plugin installed, the skill is invoked as /kapacitor:errors in Claude Code; in Codex and Cursor it’s the kapacitor-errors skill (description-matched, so the agent picks it up from a natural question — you don’t normally type the name).

Or ask naturally:

Show me the failed tool calls from the last session.

kapacitor validate-plan

Section titled “kapacitor validate-plan”

Verify every item in a session’s plan was completed.

Terminal window
kapacitor validate-plan <sessionId>

The command reads the plan (extracted from the agent’s tool calls — Claude Code’s TodoWrite, or a markdown plan committed during the session) and checks each item against what actually happened. Output is a list of completed / pending / unverified items.

Inside the agent: /kapacitor:validate-plan in Claude Code, or the kapacitor-validate-plan skill in Codex and Cursor.

Or:

Did I finish everything in the plan? Check what's left to do.

This is most useful at the end of a multi-step session, before claiming a task is done.

Section titled “Related”
  • Session recap — the narrative summary.
  • Evaluations — structured scoring across 13 quality dimensions, including a dedicated “repeated failed attempts” judge.