← Back to Skills Marketplace
spyfree

ACPX Codex Playbook

by Li Xin · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
339
Downloads
0
Stars
1
Active Installs
1
Versions
Install in OpenClaw
/install acpx-codex-playbook
Description
Practical playbook for running Codex through acpx in persistent sessions, especially when the task needs reliable file creation, local dependency installs, s...
README (SKILL.md)

acpx-codex-playbook

Use acpx as a structured control plane for Codex. Prefer persistent sessions, prompt files, and shell-based file generation over fragile one-shot prompts and tool-native file writes.

Quick start

Run this default flow for any non-trivial task:

acpx codex sessions new --name task
acpx codex set-mode -s task full-access
acpx codex -s task -f prompt.txt

Prefer this over acpx codex exec ... when the task needs iteration, file output, validation, or retries.

Workflow

1. Choose session type

Use exec only for small one-shot tasks.

Use a persistent session when the task involves any of the following:

  • generating files
  • multiple retries
  • long prompts
  • validation steps
  • local installs or virtual environments
  • deliverables such as .pptx, .docx, reports, videos, or scripts

2. Set mode explicitly

For practical work, set mode before prompting:

acpx codex set-mode -s task full-access

Interpretation:

  • read-only: inspect only
  • auto: moderate default behavior
  • full-access: broader session capability, including easier file edits and broader path/network freedom

Do not assume full-access means sudo or root. It relaxes the ACP session; it does not guarantee system-level privilege escalation.

3. Use prompt files, not huge shell strings

For long or delicate instructions, always write a prompt file and pass -f:

cat > prompt.txt \x3C\x3C'TXT'
Task: ...
Constraints: ...
Outputs: ...
Validation: ...
TXT

acpx codex -s task -f prompt.txt

This avoids shell quoting failures and makes retries reproducible.

4. Prefer shell/Python file writes over ACP fs writes

If the task must create or rewrite files, instruct Codex to prefer:

  • shell heredocs
  • python - \x3C\x3C'PY' ... PY
  • direct command-line generation

Prefer these over tool-native fs/write_text_file style edits when prior attempts showed permission failures.

Recommended instruction snippet:

If built-in file-editing tools fail, write files via shell heredoc or Python scripts instead of ACP fs write calls.

5. Use writable output paths first

For fragile generation tasks, write outputs under /tmp first, validate them, then move/copy them into the target workspace.

Recommended pattern:

  • generate under /tmp/...
  • validate structure and existence
  • copy to final destination only after success

This is especially useful for generated binaries like .pptx.

6. Validate before declaring success

Always ask Codex to verify outputs.

Examples:

  • file exists
  • zip/XML structure parses for .pptx
  • image dimensions or PDF page count
  • report file with output path, validation result, and model if visible

Practical patterns

Pattern: generated deliverables

For PPT/report/document generation, require all of the following in the prompt:

  • exact output path
  • exact report path
  • validation steps
  • final two-line summary with model/path if possible

See references/ppt-playbook.md for a concrete template.

Pattern: local dependency installs

If non-stdlib packages are needed, prefer project-local installs:

python3 -m venv .venv
. .venv/bin/activate
pip install \x3Cpackage>

Avoid assuming global install rights. Use system-level installs only when explicitly intended and actually permitted by the host.

Pattern: troubleshooting failed writes

If touch or shell writes work but ACP file edits fail, treat it as an ACP handler or sandbox-path issue, not proof that Codex itself lacks capability. Switch the generation strategy to shell/Python writes.

Decision rules

  • If the task is small and read-heavy: exec is acceptable.
  • If the task must create deliverables: use a persistent session.
  • If the prompt is long: use -f prompt.txt.
  • If file editing fails once: switch to shell/Python write strategy.
  • If dependencies are missing: try local .venv install before changing system state.
  • If a binary artifact is required: generate in /tmp, validate, then move.

Anti-patterns

Avoid these common failure modes:

  • stuffing long multilingual prompts directly into one shell string
  • assuming full-access equals sudo/root
  • relying only on ACP fs writes for large generated files
  • declaring success before validating output structure
  • writing the final artifact directly into a path that may be sandbox-restricted

References

  • Read references/ppt-playbook.md when the task is to generate a PPT or similar structured binary deliverable via acpx/Codex.
  • Read references/troubleshooting.md when acpx sessions start but file creation, mode behavior, or sandbox boundaries are unclear.
Usage Guidance
This is a how-to playbook and appears internally consistent. Before using it, understand that following its advice will cause the agent (or you) to run shell commands, create files under /tmp, and potentially install Python packages into a local venv — these actions modify system state and may require network access for pip installs. If you are concerned about safety: run the workflow in an isolated/test environment, avoid placing secrets into prompt files, review generated commands before execution (or run in read-only/exec mode for verification), and validate artifacts in /tmp before moving them to persistent locations. The skill does not request credentials or hidden network endpoints.
Capability Analysis
Type: OpenClaw Skill Name: acpx-codex-playbook Version: 1.0.0 The skill bundle provides a playbook for using the 'acpx' tool to execute code and manage files, explicitly encouraging the use of 'full-access' modes and shell-based bypasses for sandbox or permission restrictions (SKILL.md, references/troubleshooting.md). While these instructions facilitate arbitrary code execution and file system manipulation, they appear intended for task automation and reliability rather than clear malicious exploitation or data exfiltration.
Capability Assessment
Purpose & Capability
Name/description (acpx + Codex persistent sessions, file creation, local installs, deliverables) align with the actual content: only runtime guidance for using acpx, using /tmp, using venvs, and preferring shell/Python file writes. There are no unrelated env variables, binaries, or config paths required.
Instruction Scope
The SKILL.md explicitly instructs the agent/operator to set session mode to 'full-access', run shell heredocs, create files under /tmp, and prefer project-local pip installs in virtualenvs. These are coherent with the playbook's purpose but mean the agent (or operator following it) will run shell commands and may install packages — actions that change system state. The instructions do not direct reading or exfiltrating unrelated system secrets or contacting unexpected external endpoints.
Install Mechanism
No install spec is present (instruction-only), so nothing is written to disk by the skill itself. The playbook recommends runtime actions (creating a .venv and pip installing) which are normal operational guidance; there is no packaged download URL or extract step in the skill metadata to raise additional install risk.
Credentials
The skill requests no environment variables, credentials, or config paths. Recommendations to use /tmp, local .venv, and 'full-access' are proportional to the stated goal of producing local artifacts and resolving filesystem/sandbox issues.
Persistence & Privilege
The skill is not always-enabled and is user-invocable (defaults). It does recommend setting an acpx session to 'full-access' during operation; while that expands session-level capabilities, the skill itself does not request persistent platform-level privileges or modify other skills or system-wide agent settings.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install acpx-codex-playbook
  3. After installation, invoke the skill by name or use /acpx-codex-playbook
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release: practical acpx+Codex workflow, prompt-file pattern, full-access guidance, shell-write fallback, troubleshooting, and PPT generation playbook.
Metadata
Slug acpx-codex-playbook
Version 1.0.0
License MIT-0
All-time Installs 1
Active Installs 1
Total Versions 1
Frequently Asked Questions

What is ACPX Codex Playbook?

Practical playbook for running Codex through acpx in persistent sessions, especially when the task needs reliable file creation, local dependency installs, s... It is an AI Agent Skill for Claude Code / OpenClaw, with 339 downloads so far.

How do I install ACPX Codex Playbook?

Run "/install acpx-codex-playbook" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is ACPX Codex Playbook free?

Yes, ACPX Codex Playbook is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does ACPX Codex Playbook support?

ACPX Codex Playbook is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created ACPX Codex Playbook?

It is built and maintained by Li Xin (@spyfree); the current version is v1.0.0.

💬 Comments