Errors & plan validation

Use kcap errors to find failures, and kcap validate-plan to collect the plan and evidence needed to assess completion.

Find tool errors

kcap errors <sessionId>
kcap errors <meta-session-slug>
kcap errors --chain <sessionId>

The output groups recorded tool errors so you can inspect failed commands, file operations, and repeated attempts. Use --chain when the work spans continuations.

An error is evidence to investigate. It may have been recovered later in the session; read the surrounding transcript before treating it as unfinished work.

From your agent, ask:

Show me the failed tool calls from this session and check which ones were resolved.

Validate a plan

kcap validate-plan <sessionId>

In v1.1.1, this command presents declared or discovered plan artifacts, the task ledger and its statuses, available outcome summaries, and recorded work. It supplies comparison instructions so you or your agent can assess the evidence.

Ask:

Check every item in the plan against what we actually did. Identify anything incomplete or unverified.

A task marked completed is an agent declaration. Verify it against the changes and relevant checks before concluding that the plan is finished.

Missing or incomplete evidence

The output marks unavailable artifact content, truncated text, and unresolved newer revisions. Do not treat the last available revision as proof that the latest plan was satisfied.

For scripts, exit codes describe whether the evidence could be rendered:

Exit code	Meaning
0	Normal output, including “No plan found.”
1	A server or access error.
2	The leading plan artifact’s content is unavailable.

Exit code zero is not a completion verdict.

Record the plan while working

The installed plans skill uses the kcap-plans MCP tools to declare documents and keep task statuses current. That record appears with session outcomes and survives session continuation.

Use Evals for broader quality assessment, or Session recap and recall to read the work’s narrative and turns.