chatgpt-chat

Imported from trchopan/openlia (profile/skills/chatgpt-chat/SKILL.md). Install upstream with npx skills add trchopan/openlia --skill chatgpt-chat. Copyright stays with the author.

ChatGPT Chat (Automated Script Execution)

When to Use

Use this skill to leverage the ChatGPT web interface (including ChatGPT Search, Advanced Data Analysis, or thinking models) to research, draft, compare technical architectures, or offload complex inquiries, while guaranteeing user privacy through Temporary Chat mode and saving the complete multi-turn dialogue into structured YAML.

• Deep Reasoning & Comparison: Offload complex comparative analysis or code generation to specialized ChatGPT models.
• Privacy-Preserving Research: Enforce Temporary Chat mode so discussions are neither saved to user chat history nor used for model training.
• Knowledge Base Ingestion: Serialize verbatim turns into standard YAML containing rich Markdown for durable storage in workspace/knowledge/chatgpt/.

Strict Safety & Gatekeeper Rules

[!IMPORTANT] Zero-Tolerance Temporary Chat Policy:

• NEVER post queries into a standard, persistent chat session.
• The automated script explicitly verifies that Temporary Chat mode is active before typing or submitting any prompt.
• If Temporary Chat cannot be confirmed, the script HALTS IMMEDIATELY, resets to a blank page, and raises an error.

Agent Operational Procedures

Primary Execution: Dedicated Automation Script

To query ChatGPT, run the bundled script via terminal:

python3 /opt/data/skills/browser-pilot/scripts/openlia_job.py submit chatgpt-chat \
  --prompt "Explain the key differences between Go channels and Rust channels with code examples." \
  --topic "Go vs Rust Channels"

The command returns a job ID immediately. Poll it with status JOB_ID and fetch the completed YAML with result JOB_ID; do not hold one terminal invocation open for the full remote-browser conversation.

What the Script Handles Automatically:

1. MCP Connection: Connects to the local/remote Playwright MCP server (localhost:8931 or OPENLIA_BROWSER_MCP_URL).
2. Temporary Navigation: Navigates to https://chatgpt.com/?temporary-chat=true.
3. Gatekeeper Verification: Verifies the "Temporary Chat" badge or "Chat history is turned off" text in the accessibility tree before typing. Fails closed if unverified.
4. Prompt Submission: Inputs text into #prompt-textarea and clicks the send button (or presses Enter).
5. Streaming Completion: Monitors generation until streaming finishes, the stop button disappears, and the send button is re-enabled.
6. Authentic Markdown & Citation Extraction: Intercepts navigator.clipboard.writeText and triggers ChatGPT's "Copy response" button to capture 100% authentic Markdown (tables, code blocks, headers), cleans tracking query parameters, and derives clean citation titles.
7. YAML Serialization: Writes the conversation to workspace/knowledge/chatgpt/YYYYMMDD_HHMMSS_<slug>.yaml.
8. Session Cleanup: Resets the browser tab to about:blank.

Multi-Turn Conversation (Follow-ups)

To continue a conversation without resetting the chat session:

python3 /opt/data/skills/browser-pilot/scripts/openlia_job.py submit chatgpt-chat \
  --prompt "Now compare performance and memory overhead." \
  --topic "Go vs Rust Channels follow-up"

Developer Verification Workflow

Before deploying skill updates to a live agent, developers must verify skill health:

1. Offline Self-Test (Fast / CI)

python3 profile/skills/chatgpt-chat/scripts/chatgpt_conversation.py --self-test

Validates URL cleaning, tracking parameter stripping, title cleaning, site metadata inference, and YAML schema compliance with zero external dependencies.

2. Live Browser Verification (Local Playwright MCP)

python3 profile/skills/chatgpt-chat/scripts/chatgpt_conversation.py --verify

Executes a canary query against the running local Playwright MCP, saves isolated test output to /tmp/openlia_verify/chatgpt/, validates schema, and reports pass/fail without contaminating user workspace data.

UI Selector Reference