All skills
Skillintermediate

pal codereview invocation

`pal codereview` is a **multi-turn** tool. Step 1 outlines the review strategy; the expert analysis that produces actual findings runs as a follow-up. A single-step call returns zero findings.

Claude Code Knowledge Pack7/10/2026

Overview

pal codereview invocation

pal codereview is a multi-turn tool. Step 1 outlines the review strategy; the expert analysis that produces actual findings runs as a follow-up. A single-step call returns zero findings.

Step 1 — initial call

Use these parameters:

ParameterValue
modelThe most capable model from pal listmodels
stepThe content to review (document text or git diff), prefixed with a framing instruction
step_number1
total_steps2
next_step_requiredtrue
review_validation_type"external" (enables expert follow-up that produces findings)
thinking_mode"max"
review_type"full"
findings"Initial submission for review. No findings yet."
relevant_filesAbsolute paths of the files being reviewed
confidence"exploring"

Step 2 — continuation call

After step 1 returns, make a follow-up call using the continuation_id from the step 1 response:

ParameterValue
modelSame model as step 1
continuation_idFrom step 1 response
step"Produce the expert analysis and final findings based on the review in step 1."
step_number2
total_steps2
next_step_requiredfalse
findingsCopy findings from step 1 response (or summarize if too large)
confidence"high"

Parallel execution with sub-agents

The step 1 call can be issued in the same message as the sub-agent (Agent tool) call — they execute in parallel. When both return, make the pal step 2 continuation call. The sub-agent typically takes longer than pal step 1, so the continuation call adds minimal wall-clock time.

Failure modes

  • listmodels returns no models → skip pal, proceed with sub-agent findings only
  • Step 1 succeeds but step 2 fails → use any findings from step 1 response
  • Both steps return zero issues → treat as a soft failure (pal produced no signal); note in the report and proceed with sub-agent findings

pal is an external model with no conversation context — naturally isolated. Its output stays in native format — do NOT map it to the skill's finding types or severity levels.