From ai-scrum
External review via Codex CLI (cross-model second opinion). Activated if codex-plugin-cc installed. Falls back gracefully if Codex unavailable. Provides VERDICT: AGREE / DISAGREE / FOUND_NEW_ISSUES.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ai-scrum:codex-external-reviewerThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Cross-model review через Codex CLI - independent second opinion.
Cross-model review через Codex CLI - independent second opinion.
Check codex-plugin-cc:
# Check if /codex:review command available (indicates plugin installed)
if ! command_available "/codex:review"; then
log "codex-plugin-cc not installed - skip external review"
return SKIP
fi
# Check Codex CLI working
if ! codex --version >/dev/null 2>&1; then
log "Codex CLI not responding - cannot do external review"
return SKIP_TECHNICAL
fi
If Codex unavailable - graceful skip:
Когда Codex available:
Sprint context:
# Sprint {{N}} for External Review
## Goal
{{GOAL}}
## Tasks
{{TASKS_SUMMARY}}
## Increment
Files created/modified в этом sprint:
{{FILES_LIST}}
## Internal SE review findings
{{SE_FINDINGS}}
## Question for Codex
Do you agree with internal review? Have any new issues been missed?
Apply adversarial review - look for what Claude might have structurally missed.
Saved to /tmp/codex-context-{UUID}.md.
Before sending to Codex - check for secrets:
# Use gitleaks или similar to scan context
gitleaks detect --source /tmp/codex-context-${UUID}.md --no-git
# If secrets found - redact before sending
if secrets_found; then
redact_secrets /tmp/codex-context-${UUID}.md
fi
Через codex-plugin-cc command:
# Use existing plugin command
/codex:adversarial-review "$(cat /tmp/codex-context-${UUID}.md)"
OR direct codex CLI invocation:
codex exec -m gpt-5.4-codex -s read-only \
-o /tmp/codex-result-${UUID}.md \
"$(cat /tmp/codex-context-${UUID}.md)"
ВАЖНО: -s read-only - Codex auditor никогда modifies code.
Codex output должен content match:
If output не conforms - retry once с reformulated prompt. If still bad - log error, mark review as FAILED_TECHNICAL.
.sprints/active/reviews/codex-review-sprint-{N}.md:
# Codex External Review: Sprint {{N}}
**Reviewed by**: Codex (gpt-5.4-codex via codex-plugin-cc)
**Reviewed at**: {{TIMESTAMP}}
## Verdict
{{AGREE_WITH_CLAUDE | DISAGREE | FOUND_NEW_ISSUES}}
## Findings (if any)
{{FINDINGS_LIST}}
## Comparison with internal SE review
{{COMPARISON}}
## Action items
{{ACTIONS}}
Skip review with logging:
DeepSeek fallback (если configured):
reviews.external.fallback.enabled: truereviewed_by: deepseek-v4 для transparency# Fallback invocation
aider --model openrouter/deepseek/deepseek-v4-pro \
--no-auto-commits --yes-always \
--message-file /tmp/codex-context-${UUID}.md \
> /tmp/fallback-result-${UUID}.md 2>&1
If user in restricted geography:
Before any send:
Codex CLI invoked с -s read-only:
Все Codex invocations logged:
.sprints/active/reviews/codex-review-sprint-{N}.md if review ran.
Updated _state.json:
external_review_status: passed | failed | skipped | unavailableexternal_review_verdict: AGREE | DISAGREE | FOUND_NEW_ISSUES (if completed)When Codex DISAGREES with internal SE - human judgment required, не auto-revert.
npx claudepluginhub shakhovskiya-create/shakhoff-claude-marketplace --plugin ai-scrumGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.