Collaborating With Claude Agent Skill

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>.

Quick start (shell-safe)

⚠️ If your prompt contains Markdown backticks ( like/this ), do not pass it directly via --PROMPT "..." (your shell may treat backticks as command substitution). Use a heredoc instead; see references/shell-quoting.md.

PROMPT="$(cat <<'EOF'
Review src/auth.py around login() and propose fixes.
OUTPUT: Unified Diff Patch ONLY.
EOF
)"
python3 .codex/skills/collaborating-with-claude/scripts/claude_bridge.py --cd "." --model sonnet --PROMPT "$PROMPT" --output-format stream-json

Output: JSON with success, SESSION_ID, agent_messages, and optional error / all_messages.

Multi-turn sessions

# Start a session
PROMPT="$(cat <<'EOF'
Analyze the bug in foo(). Keep it short.
EOF
)"
python3 .codex/skills/collaborating-with-claude/scripts/claude_bridge.py --cd "." --PROMPT "$PROMPT" --output-format stream-json

# Continue the same session
PROMPT="$(cat <<'EOF'
Now propose a minimal fix as Unified Diff Patch ONLY.
EOF
)"
python3 .codex/skills/collaborating-with-claude/scripts/claude_bridge.py --cd "." --SESSION_ID "<SESSION_ID>" --PROMPT "$PROMPT" --output-format stream-json

Prompting patterns (token efficient)

Use assets/prompt-template.md as a starter when crafting --PROMPT.

1) Ask Claude to open files itself

Provide:

• Entry file(s) and approximate line numbers
• Objective and constraints
• Output format (diff vs analysis)

Avoid:

• Pasting large code blocks
• Multiple competing objectives in one request

2) Enforce safe output for code changes

Append this to prompts when requesting code:

• OUTPUT: Unified Diff Patch ONLY. Strictly prohibit any actual modifications.

3) Use Claude for what it’s good at

• Alternative solution paths and edge cases
• UI/UX and readability feedback
• Review of a proposed patch (risk spotting, missing tests)

Verification

• Smoke-test the bridge: python3 .codex/skills/collaborating-with-claude/scripts/claude_bridge.py --help.
• If you need a session: run one prompt with --output-format stream-json and confirm the JSON contains success: true and a SESSION_ID.
• Note: --output-format text won’t include a newly generated session id; use stream-json/json to capture it. If you resume with --SESSION_ID in text mode, the bridge echoes that SESSION_ID in its JSON output.

Safety & guardrails

• Never paste secrets (private keys, API keys, seed phrases) into prompts.
• For code changes, request Unified Diff Patch ONLY and apply changes yourself.
• Treat Claude output as suggestions; verify locally (tests, lint, build) before merging.

Collaboration State Capsule

Keep this short block updated near the end of your reply while collaborating:

[Claude Collaboration Capsule]
Goal:
Claude SESSION_ID:
Files/lines handed off:
Last ask:
Claude summary:
Next ask:

References

• assets/prompt-template.md (prompt patterns)
• references/shell-quoting.md (shell quoting/backticks)