Use the Claude Code CLI to consult Claude and delegate coding tasks for prototyping, debugging, and code review. Supports multi-turn sessions via SESSION_ID. Optimized for low-token, file/line-based handoff.
SkillScore82Maintained
Automatic
◆Coding pack
Last verified 8/22/2026
What it changes
Passed 16 of 20 test cases with the skill, and 8 of 20 without it.
Without skill41%
With skill82%
Checks it passed that the model alone missed
Specifies a timeout_ms setting of 600000
Selects opus as the model flag parameter
Selects sonnet as the model flag parameter
and 6 more in the full report
9 fixed9 passed either way4 still failing0 broke
Pass-rate lift
+41 pp
Tokens
+45%
Agent turns
0%
Response time
-62%
Model gemini-3.6-flashMeasured 8/22/2026v118 of 22 cases passed with skill · 20 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 9
Prompt
I am uncertain about the supported command-line arguments for the local Claude bridge wrapper script `scripts/claude_bridge.py`. What command should be run first to inspect its interface?
What was checked, without and with the skill
Checked
Without skill
With skill
Runs python3 scripts/claude_bridge.py with the --help flag
Missed
Passed
case-12 · gemini-3.6-flash · 8/22/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md121 lines4 filesincludes scripts
Collaborating with Claude Code (Codex)
Use Claude Code CLI as a collaborator while keeping Codex as the primary implementer.
Host note: this skill is written for a Codex host. If you are already running inside Claude Code, do not bridge to Claude Code (that would be circular); use your own capabilities or another collaborator skill.
This skill provides a lightweight bridge script (scripts/claude_bridge.py) that returns structured JSON and supports multi-turn sessions via SESSION_ID.
When to use
You want a second opinion (design tradeoffs, edge cases, missing tests).
You want Claude to propose or review a unified diff (Claude does not edit files).
You want multi-turn back-and-forth while you implement locally.
When not to use
The task is trivial or one-shot (do it directly in Codex).
You need authoritative facts that require browsing/citations (Claude may guess).
You might paste sensitive data (secrets, private keys, prod logs).
Core rules
Claude is a collaborator; you own the final result and must verify changes locally.
Do not invoke claude directly; always use the bridge script (scripts/claude_bridge.py) so output/session handling stays consistent.
Prefer file/line references over pasting snippets. Run the bridge with --cd set to the repo root (it sets the claude process working directory); use --add-dir when Claude needs access to additional directories.
For code changes, request Unified Diff Patch ONLY and forbid direct file modification.
Always run the bridge script with --help first if you are unsure of parameters.
Always capture SESSION_ID and reuse it for follow-ups to keep the collaboration conversation-aware.
For automation, prefer --SESSION_ID (resume). Session selectors are mutually exclusive: choose one of --SESSION_ID, --continue, or --session-id.
Keep a short Collaboration State Capsule updated while this skill is active.
Default timeout: when invoking via the Codex command runner, set timeout_ms to 600000 (10 minutes) unless a shorter/longer timeout is explicitly required.
Default model: prefer sonnet for routine work; use opus only for complex tasks or when explicitly requested.
Ensure Claude Code is logged in before running headless commands (run claude and /login once if needed).
Streamed JSON requires --verbose; the bridge enables this automatically.
Model selection
Claude Code supports model aliases, so you can use --model sonnet / --model opus instead of hard-coding versioned model IDs.
If you omit --model, Claude Code uses its configured default (typically from ~/.claude/settings.json, optionally overridden by .claude/settings.json and .claude/settings.local.json).
If you need strict reproducibility, pass a full model name via --model <full-name>.