Help us improve
Share bugs, ideas, or general feedback.
From gaac
Calls external AI tools (Claude CLI, Codex, Gemini) in non-interactive mode for independent code review, idea proposal, design analysis, and other tasks needing fresh context.
npx claudepluginhub sihaoliu/gaac --plugin gaacHow this skill is triggered — by the user, by Claude, or both
Slash command
/gaac:third-party-callThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
This skill provides a unified interface for invoking external AI tools in GAAC workflows. It manages:
Delegates implementation and review tasks to external AI CLI tools (Codex, Gemini) with cross-model adversarial review for cost savings and improved accuracy.
Invokes OpenAI Codex and Google Gemini CLIs via Bash for second opinions, code reviews, and alternative analysis. Useful when users request external AI verification or explicitly say 'ask codex' or 'ask gemini'.
Consults Gemini 2.5 Pro, OpenAI Codex, and Claude for second opinions on debugging failures, architectural decisions, security validation, and fresh perspectives.
Share bugs, ideas, or general feedback.
This skill provides a unified interface for invoking external AI tools in GAAC workflows. It manages:
External tools provide:
| Tool | Primary Use | Detection |
|---|---|---|
codex | Code review, analysis | which codex |
claude | Fallback for codex | which claude |
gemini | Web research, proposals | which gemini |
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/check-tools.sh"
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-analysis.sh" \
--prompt-file ./prompt.md \
--output-file ./output.md \
--tool codex \
--context-files "src/**/*.ts"
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-web-research.sh" \
--topic "React state management best practices 2024" \
--output-file ./research.md
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-peer-check.sh" \
--issue-number 42 \
--output-file ./peer-review.md
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-code-review.sh" \
--issue-number 42 \
--output-file ./code-review.md
Outputs structured markers: <!-- GAAC_REVIEW_SCORE: NN --> and <!-- GAAC_REVIEW_ASSESSMENT: ... -->
Model configuration is read from gaac.md. The --role parameter determines which model config to use:
| Role | gaac.md Key | Default | Purpose |
|---|---|---|---|
analyzer | gaac.models.analyzer | codex:gpt-5.2-codex:high | Synthesis and analysis |
analyzer (fallback) | gaac.models.analyzer_fallback | claude:opus | Fallback for analyzer |
checker | gaac.models.checker | claude:opus | Critical checking |
proposer | gaac.models.proposer | claude:sonnet | Creative proposals |
proposer_secondary | gaac.models.proposer_secondary | gemini:gemini-3-pro-preview | Secondary proposer (Gemini) |
code_reviewer | gaac.models.code_reviewer | codex:gpt-5.2-codex:high | Independent code review |
code_reviewer (fallback) | gaac.models.code_reviewer_fallback | claude:opus | Fallback for code review |
Usage:
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-analysis.sh" \
--role analyzer \
--prompt-file ./prompt.md \
--output-file ./output.md
Model Config Format: tool:model:reasoning (e.g., codex:gpt-5.2-codex:high)
| Prompt | Purpose |
|---|---|
prompts/CODE_REVIEWER_PROMPT.md | 100-point scoring rubric for code review |
prompts/CHECKER_PROMPT.md | Critical checker for proposer-checker-analyzer debates |
codex exec \
-m gpt-5.2-codex \
-c model_reasoning_effort=high \
--enable web_search_request \
-s read-only \
-o "$OUTPUT_FILE" \
-C "$PROJECT_ROOT" \
< prompt.txt
Features:
high effort)claude -p \
--model opus \
--permission-mode bypassPermissions \
--tools "Read,Grep,Glob,WebSearch,WebFetch" \
--allowedTools "Read,Grep,Glob,WebSearch,WebFetch" \
< prompt.txt > output.md
Features:
GEMINI_CLI_SYSTEM_SETTINGS_PATH="$TEMP_SETTINGS" \
gemini \
--output-format json \
--approval-mode yolo \
"prompt" > output.json
Settings file:
{
"tools": {
"core": ["list_directory", "read_file", "glob", "search_file_content", "web_fetch", "google_web_search"],
"exclude": ["write_file", "replace", "run_shell_command"]
}
}
Features:
All external tools run with:
| Capability | Allowed | Enforced By |
|---|---|---|
| Read codebase | Yes | Tool config |
| Search codebase | Yes | Tool config |
| Web search | Yes | Tool config |
| Write files | No | Sandbox/tool exclusion |
| Execute commands | No | Sandbox/tool exclusion |
Uses Codex (or Claude fallback) to review implementation:
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-peer-check.sh" --issue-number 42
Output: PASS/NEEDS_WORK with detailed findings
Uses Gemini for web-enhanced proposals:
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-web-research.sh" \
--topic "distributed cache invalidation patterns" \
--output-file ./proposal.md
Uses Codex/Claude for synthesizing multiple inputs:
bash "${CLAUDE_PLUGIN_ROOT}/skills/third-party-call/scripts/run-analysis.sh" \
--prompt-file ./analysis-prompt.md \
--context-files "./docs/draft/*.md" \
--output-file ./synthesis.md
| Exit Code | Meaning | Action |
|---|---|---|
| 0 | Success | Use output |
| 1 | Tool not available | Try fallback |
| 2 | Tool execution failed | Try fallback |
| 3 | Invalid input | Fix and retry |
| 4 | Timeout | Increase timeout or simplify |
| Tool | Default | Env Variable |
|---|---|---|
| Codex | 5400s | CODEX_TIMEOUT |
| Claude | 300s | CLAUDE_TIMEOUT |
| Gemini | 300s | GEMINI_TIMEOUT |
| GAAC Stage | Third-Party Usage |
|---|---|
| Research | Gemini for web proposals |
| Refinement | Codex/Claude for analysis |
| Planning | Codex for design review |
| Work | Peer-check for code review |
| File | Purpose |
|---|---|
SKILL.md | This documentation |
scripts/check-tools.sh | Detect available tools |
scripts/run-analysis.sh | General analysis wrapper with role-based config |
scripts/run-web-research.sh | Gemini web research |
scripts/run-peer-check.sh | Quick feedback via external model |
scripts/run-code-review.sh | Independent scoring review (mandatory) |
prompts/CODE_REVIEWER_PROMPT.md | 100-point scoring rubric |
prompts/CHECKER_PROMPT.md | Critical checker for debates |