cursor:review - Advisory Cursor Review
SkillAI & modelsLets your agent get a second-opinion code review from Cursor Composer while keeping the final verdict with your main agent.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the cursor:review - Advisory Cursor Review skill
About this capability
Run a Cursor Composer review as an advisory second opinion while keeping the primary review verdict on the host brain. Use when user invokes cursor:review, asks Cursor to review, or wants composer to sanity-check a diff. Cursor never owns APPROVE/REQUEST_CHANGES.
What this skill tells your AI
The instructions your AI receives, as published by chachamaru127/claude-code-harness in skills/cursor-review/SKILL.md and read by ahel’s review.
Cursor Composer (composer-2.5-fast) を read-only second opinion として使う review skill。primary verdict は host brain (Opus / Claude role) が出す。
Quick Reference
cursor:review --base origin/main
cursor:review "Phase 88.5 command namespace diff"
Rules
-
Cursor review is advisory. The final verdict must come from the host reviewer.
-
Do not pass
--writetocursor-companion.sh. -
Use resolver/model routing, not direct env checks:
bash scripts/resolve-impl-backend.sh --backend cursor --role reviewer bash scripts/model-routing.sh --host cursor --role worker --field model -
Prefer the existing
harness-reviewcontract when a full review is requested. This command exists for users who explicitly ask for the Cursor lane.
Flow
-
Resolve the Harness helper root:
HARNESS_PLUGIN_ROOT="${HARNESS_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-}}" if [ -z "$HARNESS_PLUGIN_ROOT" ] && [ -n "${CLAUDE_SKILL_DIR:-}" ]; then probe="$(cd "${CLAUDE_SKILL_DIR}" && pwd)" while [ "$probe" != "/" ] && [ ! -d "$probe/scripts" ]; do probe="$(cd "$probe/.." && pwd)" done [ -d "$probe/scripts" ] && HARNESS_PLUGIN_ROOT="$probe" fi if [ -z "$HARNESS_PLUGIN_ROOT" ]; then echo "ERROR: HARNESS_PLUGIN_ROOT is not set and could not be derived from CLAUDE_PLUGIN_ROOT or CLAUDE_SKILL_DIR" >&2 exit 2 fi -
Determine review scope:
BASE_REF="${TASK_BASE_REF:-origin/main}" REVIEW_SCOPE="${ARGUMENTS:-}" if [ -n "${REVIEW_SCOPE}" ]; then set -- ${REVIEW_SCOPE} while [ "$#" -gt 0 ]; do case "$1" in --base) BASE_REF="${2:?--base requires a ref}" shift 2 ;; --base=*) BASE_REF="${1#--base=}" shift ;; *) shift ;; esac done fi DIFF_STAT="$(git diff --stat "${BASE_REF}..HEAD")" DIFF_TEXT="$(git diff "${BASE_REF}..HEAD")" -
Ask Cursor in read-only mode:
PROMPT="$(cat <<EOF
Review this diff as an advisory second opinion. Base ref: ${BASE_REF} Requested scope: ${ARGUMENTS:-full diff}
Focus on bugs, regressions, missing tests, and unsafe assumptions. Do not propose broad refactors.
Diff stat: ${DIFF_STAT}
Diff: ${DIFF_TEXT} EOF )" bash "${HARNESS_PLUGIN_ROOT}/scripts/cursor-companion.sh" task "${PROMPT}"
4. Host reads Cursor's output and performs the primary review:
- Findings first, ordered by severity.
- Mark Cursor-originated points as advisory when used.
- `APPROVE` / `REQUEST_CHANGES` must be the host decision, not a copied Cursor verdict.
## Output
Use the standard review shape:
- findings
- open questions
- validation run
- final host verdict
Signals
- GitHub stars
- 3k
- Forks
- 299
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
cursor-review- Source
- github.com/chachamaru127/claude-code-harness