responses-third-party-prompt-cache-patch

A GitHub-ready wrapper repo for the responses-third-party-prompt-cache-patch OpenClaw skill.

This skill applies a local patch to an installed OpenClaw dist bundle so third-party OpenAI-compatible Responses endpoints keep prompt_cache_key and prompt_cache_retention instead of having them stripped.

中文版本：README.zh-TW.md

ClawHub: responses-third-party-prompt-cache-patch

Quick install for OpenClaw

Install the skill into the default OpenClaw workspace:

clawhub install responses-third-party-prompt-cache-patch --workdir ~/.openclaw/workspace

After install, follow the apply workflow below or use the OpenClaw paste-in instruction.

Paste this into OpenClaw

Use this one-liner if you want OpenClaw to fetch the install guide and carry out the install/apply workflow on the target machine:

Fetch and follow instructions from https://raw.githubusercontent.com/tsunheimat/openclaw-responses-prompt-cache-patch/refs/heads/main/.openclaw/INSTALL.md

Use the raw URL, not the GitHub blob/... page. The .openclaw/INSTALL.md file is written as an OpenClaw-specific execution guide: install the skill, run dry-run, apply the patch, restart the gateway, and report the result.

Who this is for

Use this if you:

• run OpenClaw on your own machine or server,
• use a third-party OpenAI-compatible Responses endpoint,
• need prompt-cache hints to survive the OpenClaw bundle layer,
• are okay applying a local patch to the installed OpenClaw dist/ files.

Do not use this if you want an upstream OpenClaw code change, a config-only fix, or a published marketplace install flow.

What this repo contains

.
├── .openclaw/INSTALL.md
├── .gitignore
├── README.md
└── skill/
    └── responses-third-party-prompt-cache-patch/
        ├── SKILL.md
        └── scripts/
            ├── _bundle_patch_common.py
            ├── patch_prompt_cache.py
            └── revert_prompt_cache.py

The public-facing repo docs live at the repo root. The installable skill stays clean under skill/responses-third-party-prompt-cache-patch/ so it can still be packaged or distributed separately.

Additional docs

• Technical walkthrough: docs/TECHNICAL-OVERVIEW.md
• Disclaimer: docs/DISCLAIMER.md

Important risks

• The apply script writes directly into the installed OpenClaw dist/ directory.
• The patch is local, not upstream. OpenClaw upgrades or reinstalls may require you to apply it again.
• You must restart the gateway after apply or rollback:

  openclaw gateway restart

• Installing the skill folder alone does not activate the patch. You must run the apply script.

Install onto another OpenClaw machine

1. Get a local copy of this repo onto the target machine. If you are viewing it on GitHub, use Code → Download ZIP, clone with the real URL shown on the page, or copy an existing checkout with scp / rsync.
2. From the cloned or unpacked repo root, copy the skill into the OpenClaw workspace skills directory:

   mkdir -p ~/.openclaw/workspace/skills
   cp -R ./skill/responses-third-party-prompt-cache-patch ~/.openclaw/workspace/skills/

3. Move into the installed skill directory:

   cd ~/.openclaw/workspace/skills/responses-third-party-prompt-cache-patch

4. Dry-run first:

   python3 scripts/patch_prompt_cache.py --dry-run

5. Apply the patch:

   python3 scripts/patch_prompt_cache.py

6. Restart OpenClaw Gateway:

   openclaw gateway restart

For the OpenClaw-specific install/apply playbook, see .openclaw/INSTALL.md.

Direct raw URL: https://raw.githubusercontent.com/tsunheimat/openclaw-responses-prompt-cache-patch/refs/heads/main/.openclaw/INSTALL.md

Apply / rollback commands

Run these from the installed skill directory:

cd ~/.openclaw/workspace/skills/responses-third-party-prompt-cache-patch
python3 scripts/patch_prompt_cache.py --dry-run
python3 scripts/patch_prompt_cache.py
openclaw gateway restart

Rollback:

cd ~/.openclaw/workspace/skills/responses-third-party-prompt-cache-patch
python3 scripts/revert_prompt_cache.py
openclaw gateway restart

Notes for distribution

• This repo is meant to be a clean GitHub distribution wrapper.
• If the skill is later published to ClawHub, that can become an additional install path, but this repo remains a manual fallback and source-of-truth-friendly distribution channel.
• The patch and rollback logic are intentionally kept in the skill scripts; this repo does not change their behavior.