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.
npx agentmods add instructions/sorryorc/mcp-codex-worker/agents-mdgit clone --depth 1 https://github.com/sorryorc/mcp-codex-workerWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00389 | $0.00389 |
| Opus 5 | $0.00195 | $0.00195 |
| Sonnet 5 | $0.00078 | $0.00078 |
| Haiku 4.5 | $0.00039 | $0.00039 |
Grade A, and why
mcp-codex-worker AGENTS.md scanned grade A with 0 findings 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 yesterday.
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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
AGENTS.md - Codex Worker Routing Rules
This repository exists to reduce premium Codex context usage. Treat worker routing as part of the task, not as an optional optimization.
Routing Contract
- Before repository-wide search, bulk file reads, review, diffs, test output, or
implementation loops, expose
mcp__codex_async_workertools withtool_searchif they are not already callable. - Use
mcp__codex_async_worker.searchfor repo discovery instead of main-threadrgwhen the output could be more than a few lines. - Use
mcp__codex_async_worker.read_packoranalyzefor file-reading and explanation tasks instead of reading whole files into the main thread. - Use
mcp__codex_async_worker.shellwithdigest:truefor tests, builds, lint, and noisy commands. - Use
mcp__codex_async_worker.diff_digestandreviewbefore ingesting full diffs. - Use
mcp__codex_async_worker.startfor implementation loops. Preferinclude_diff:false, scoped paths, and explicit checks.
Main-Thread Budget
Keep the main Codex thread for planning, decisions, small verification, and final review. Do not use main-thread shell or file reads as a substitute for worker tools when the task is large enough to affect context usage.
Audits
- Run
npm run codex:audit -- --since-minutes=60when checking whether recent work is being routed through the worker. - Run
npm run stats:gatefor the existing metrics gate. - Remember that worker metrics can only see worker/tool audit rows. Direct main-thread file reads, shell output, and chat context are outside the metrics file unless explicitly instrumented.
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.
- yesterday First seen · 36 lines · 389 tokens per session scan A b11b3cd8904f
mcp-codex-worker AGENTS.md is an instructions file published in the GitHub repository sorryorc/mcp-codex-worker (2 stars, last pushed 14d ago), licensed MIT. It adds 389 tokens to every session, about $0.0019 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
claude-cost-optimizer CLAUDE.md
Instructions for Sagargupta16/claude-cost-optimizer, covering claude.md - claude-cost-optimizer, purpose, file structure, writing conventions and python tools.
ClawRouter CLAUDE.md
Claude Code instructions for BlockRunAI/ClawRouter, covering clawrouter, commands, project structure, key dependencies and conventions.
aider-desk AGENTS.md
AGENTS.md instructions for hotovo/aider-desk, covering agents.md, common commands, development, type checking and linting and formatting.
Meta_Kim CLAUDE.md
Instructions for KimYx0207/Meta_Kim, covering metakim for claude code, fast read, what this repository is, claude code's role and canonical vs runtime files.
crystal CLAUDE.md
Instructions for stravu/crystal, covering crystal - multi-session claude code manager, project overview, references, implementation status: ✅ complete and ✅ implemented features.
easy-cheese python.instructions.md
Instructions for paulnsorensen/easy-cheese, covering hard rules, what to flag and what not to flag.