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.
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:
| Parameter | Value |
|---|---|
model | The most capable model from pal listmodels |
step | The content to review (document text or git diff), prefixed with a framing instruction |
step_number | 1 |
total_steps | 2 |
next_step_required | true |
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_files | Absolute 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:
| Parameter | Value |
|---|---|
model | Same model as step 1 |
continuation_id | From step 1 response |
step | "Produce the expert analysis and final findings based on the review in step 1." |
step_number | 2 |
total_steps | 2 |
next_step_required | false |
findings | Copy 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
listmodelsreturns 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.