Menu
Invoke Codex CLI as a sub-agent for code tasks. Useful for second opinions, parallel exploration, or offloading isolated subtasks.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Best practices for git usage, commits, and workflow
Optimizing arithmetic circuits / constraint systems (R1CS, PLONKish, AIR) for zero-knowledge proofs — minimizing multiplication constraints, rows, witnesses, and gate degree. Use when reducing constraint/witness counts, designing or golfing R1CS/PLONK/AIR gadgets (boolean ops, adders, range checks, hashes like SHA-256/Keccak/Poseidon), doing foreign-field/non-native or CRT/RNS arithmetic, choosing lookups vs arithmetic, or using SMT (cvc5) and SageMath (Gröbner basis) to synthesize, verify, and certify constraint encodings.
Writing, proving, and debugging Lean 4 + Mathlib. Use when proving theorems, filling `sorry`s, formalizing math, fixing broken proofs, or setting up a Lean project. Covers the cached-Mathlib setup and a plan-first proving workflow.
Performance-guided Rust optimization using benchmarks, sampling profilers, and agent-readable profiling artifacts. Use when optimizing Rust throughput, latency, CPU time, allocations, code size, compile/runtime hot paths, Criterion benchmarks, hyperfine measurements, Linux perf, macOS xctrace/Instruments, samply, flamegraphs, heap profilers, cargo-bloat, cargo-llvm-lines, or cargo-asm.
The essential best practices for Rust development
Best practices for Python development
| name | codex |
| description | Invoke Codex CLI as a sub-agent for code tasks. Useful for second opinions, parallel exploration, or offloading isolated subtasks. |
| version | 1.0.0 |
Pipe the prompt on stdin; use exec for non-interactive output:
echo "<prompt>" | codex exec --skip-git-repo-check
--skip-git-repo-check lets it run outside a git repo. Drop it when invoking inside a repo where codex should see git context.
-m <model> — override model (e.g. gpt-5.1-codex).-C <dir> — set working directory.-s read-only — sandbox mode (read-only, workspace-write, danger-full-access).--dangerously-bypass-approvals-and-sandbox — fully autonomous; only inside an already-sandboxed env.-i <file> — attach image(s).--output-schema <file.json> — constrain final response to a JSON schema.Heredoc for multi-line prompts:
codex exec --skip-git-repo-check <<'EOF'
Review foo.rs for race conditions.
Report findings as a bullet list.
EOF
From a file:
codex exec --skip-git-repo-check < prompt.txt
Inside a repo, read-only second opinion:
echo "Audit auth.rs for input validation gaps." | codex exec -s read-only
run_in_background and read output later.