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 skills add gaia-research/gaia-skill-heaven --skill codex-playbookgit clone --depth 1 https://github.com/gaia-research/gaia-skill-heavenWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/gaia-research/gaia-skill-heaven/codex-playbook)<a href="https://agentmods.dev/skills/gaia-research/gaia-skill-heaven/codex-playbook"><img src="https://agentmods.dev/badge/skills/gaia-research/gaia-skill-heaven/codex-playbook/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/gaia-research/gaia-skill-heaven/codex-playbook"><img src="https://agentmods.dev/badge/skills/gaia-research/gaia-skill-heaven/codex-playbook.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00070 | $0.01882 |
| Opus 5.5 | $0.00028 | $0.00753 |
| Sonnet 5.5 | $0.00014 | $0.00376 |
| Haiku 4.5 | $0.00007 | $0.00188 |
Grade A, and why
codex-playbook scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 193 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Codex Playbook
codex is OpenAI's coding agent CLI. Its distinguishing feature for Skill Heaven is strong
isolation: config, session persistence, and the writable filesystem are each independently
suppressible from the command line, so a clean posture composes without mutating anything.
Verified against codex-cli 0.146.0. Re-check codex --version before trusting any dose
measurement.
Rule 0 — run codex in a herdr pane, never through your Bash tool. The operator must be able to see which model and reasoning effort you invoked. A probe the operator could not see is not evidence. Use
pane send-textwith a trailing newline inside the quotes — notpane run, which splits argv on whitespace and destroys quoted prompts:herdr pane send-text "$PROBE_PANE" 'codex exec --model gpt-5.6-luna -c model_reasoning_effort=low "probe text" ' herdr pane read "$PROBE_PANE"
Building the door? Read
packages/core/skills/harness-door-pattern/SKILL.mdfirst — it carries the suppression-mechanism taxonomy, the probe methodology, and the traps. This playbook is the harness-specific detail underneath it.
1. Non-interactive invocation
codex exec "your prompt"
codex exec --skip-git-repo-check --ephemeral --sandbox read-only "your prompt"
codex exec is the non-interactive entrypoint. A prompt can be passed as an argument or piped
on stdin; if both, stdin is appended as a <stdin> block.
Flags worth defaulting on for probes:
--ephemeral— do not persist session files. Always use on probes.--skip-git-repo-check— allow running outside a git repo.--sandbox read-only— the model's shell commands cannot write.
2. Model and reasoning effort
codex exec --model gpt-5.6-luna -c model_reasoning_effort=low "..."
Reasoning effort defaults to max, which is expensive — set it explicitly on every probe.
Reasoning effort is NOT a Skill Heaven posture. It sets how hard the model reasons. The Skill Heaven ladder is skill- and context-scoped — what is admitted into the session. The two are different axes that happen to share level names. Hold reasoning effort constant across benchmark arms; it is a control variable, never the thing under test.
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.
- 1mo ago First seen · 193 lines · 70 tokens per session scan A 2b6becb9fd2f
codex-playbook is a skill published in the GitHub repository gaia-research/gaia-skill-heaven (4 stars, last pushed today), licensed MIT. It adds 70 tokens to every session and 1,882 once invoked, about $0.0003 per session on Opus 5.5. 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 skills, from other repositories
architecture-compass
Architectural thinking partner for an existing repository — scans the codebase, conducts a structured interview, agrees on current architectural state and recommended direction, and produces a shareable insights document. Scoped to one repository, module, or folder. Does not execute transformation — it orients. Use…
architecture-refiner
Facilitate a structured conversation to define architecture principles for a repository. Supports multiple architecture styles: clean architecture (default), hexagonal / ports & adapters, modular monolith, or custom. Produces a formal architecture document that the corresponding atom will use. Use when setting up a…
review-refiner
Facilitate a structured conversation to customize how the review molecule works -- atom loading rules, severity classification, report format, scope rules, insight capture, and health logging. Produces a formal review-standards.md document that the review molecule will use as its process configuration. Use when the…
review
Perform a structured code review by composing validation checklists from relevant atoms based on what code changed. Loads atoms conditionally -- clean-code always, architecture/DDD/security/tests only when the delta touches their domain. Produces a severity-ordered report with specific locations and fixes. Use when…
skill-review
Deep behavioral audit of a Lattice skill — proposes 3 review personas relevant to the skill, runs independent scenario analysis from each persona's perspective, then merges only the high-confidence, practical findings into a severity-ordered gap report with proposed fixes. Structural validation (conventions…
clean-code-refiner
Facilitate a structured conversation to define clean code principles for a repository. Produces a formal clean-code.md document that the clean-code atom will use as its override. Use when setting up coding standards, defining code quality rules, or when the user says 'setup clean code', 'define coding standards'…