pair-planning

A planning process in which Claude and Codex independently plan the same task, review each other’s open questions, and agree on one implementation plan. It does not write code.

In plain words
What is it for?
Use it at the start of a coding task to compare plans, resolve disagreements, and produce a shared plan for implementation.
Why use it?
It reduces the chance that one planning perspective misses an important issue before implementation begins.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/bengous/claude-code-plugins/pair-planning
Any agent
npx skills add bengous/claude-code-plugins --skill pair-planning
Clone the repo
git clone --depth 1 https://github.com/bengous/claude-code-plugins

Made for: Claude Code, Codex.

Per session 78 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,105 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00078 $0.02105
Opus 5 $0.00039 $0.01052
Sonnet 5 $0.00016 $0.00421
Haiku 4.5 $0.00008 $0.00211

Measured 2d ago against content hash 0e7e4599f7c0, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

pair-planning scanned grade C with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

- No cleanup needed — each run uses a unique `slug`; never `rm -rf` session dirs.
agents-bridge/skills/pair-planning/SKILL.md · 161 lines

How it starts

The opening of the file, as written. The whole thing — 161 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Pair-Planning: symmetric plan-to-consensus with Codex

Two frontier models plan the same task independently, then converge. You (Claude) hold the session context and orchestrate; Codex is the fresh pair of eyes. The deliverable is a single agreed plan — read-only, no implementation.

All Codex round-trips go through the helper, never codex directly:

"${CLAUDE_PLUGIN_ROOT}/scripts/pair-codex" fresh  <dir> <prompt_file> r0     # start session
"${CLAUDE_PLUGIN_ROOT}/scripts/pair-codex" resume <dir> <prompt_file> r<n>   # continue it

The helper enforces the correct invocation (--json -o, stdin closed, read-only, thread-id capture, fail-hard). It prints Codex's final message to stdout and writes <dir>/<label>.{jsonl,last,err} plus <dir>/thread_id. Effort defaults to xhigh; override per call with CODEX_EFFORT=high when a round is light, or CODEX_EFFORT=max for a genuinely hard round. Planning is reasoning-heavy — don't downgrade the model tier.

1. Setup + launch Codex in the background

Action 0, the moment you're invoked — get Codex planning before you do, so the two plans are written in parallel and neither side anchors on the other.

  1. Pick the session slug (your handle — you control it) and create its dir:
    slug="$(date +%s)-$(printf '%s' "$ARGUMENTS" | tr -cs 'A-Za-z0-9' '-' | cut -c1-40)"
    dir="/tmp/pair-planning/$slug"; mkdir -p "$dir"; echo "$dir"
    
    Use the printed path literally in every later step — shell variables do not survive between Bash calls.
  2. Pre-flight — clarify only real blockers. Default to proceeding. Escalate with one minimal AskUserQuestion only when a genuine blocker would make the two sides plan different tasks — ambiguous objective/scope, or an irreversible high-stakes direction. Never run an upfront questionnaire. (Judgment, not a rigid gate.)
  3. Write the brief to "$dir/task.md": the user's task verbatim, needed repo context, and a short Assumptions block (scope, constraints, out of scope). Both sides plan from this identical brief, so interpretation gaps cannot resurface later as phantom OPEN-points.
  4. Write the Codex round-0 prompt to "$dir/prompt-r0.md": tell it to read "$dir/task.md", inspect the repo read-only, and produce its own independent implementation plan (files to touch, approach, risks). Anti-anchoring: round 0 points at task.md only — never reference your plan here; it does not exist for Codex yet.
  5. Launch Codex in the background so it plans while you do — run the helper with the Bash tool's run_in_background option (not blocking):
    "${CLAUDE_PLUGIN_ROOT}/scripts/pair-codex" fresh "$dir" "$dir/prompt-r0.md" r0
    
    The harness notifies you when it exits; the helper writes "$dir/r0.last" and captures Codex's thread id into "$dir/thread_id".

Read the full file on GitHub · 161 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 161 lines · 78 tokens per session scan C 0e7e4599f7c0

Subscribe to this mod's changes

pair-planning is a skill published in the GitHub repository bengous/claude-code-plugins (4 stars, last pushed 3d ago), licensed MIT. It adds 78 tokens to every session and 2,105 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.