Summary
This issue captures a series of critical friction points encountered during the initial setup and execution of VibeLab's research pipeline. The problems span across CLI authentication hangs, incorrect agent tool routing (specifically hyphen vs. underscore mismatches), inappropriate execution of skill directories, and missing environment/script dependencies.
Steps to reproduce
1.Gemini CLI Init: Attempt to log into Gemini via the VibeLab UI on a fresh environment where @google/gemini-cli is not globally installed.
2.Tool Routing: In the VibeLab Chat interface, prompt the agent with: "Please use the inno-pipeline-planner skill to start a research pipeline."
3.Claude Auth: Run the claude command in the VS Code integrated terminal to authenticate Claude Code.
4.Script Execution: Prompt the agent to execute the gemini-deep-research skill via shell commands.
Expected behavior
-
The system should prompt the user to install the missing CLI tool or handle the missing dependency gracefully.
-
The agent should correctly map the hyphenated skill name (inno-pipeline-planner) and recognize it as a conversational skill directory, initiating the interactive workflow to generate research_brief.json and tasks.json.
-
The claude CLI should successfully open a browser window for OAuth or provide a visible fallback URL in the terminal.
-
Python-based skills (like deep_research.py) should have their dependencies (e.g., requests) automatically resolved in the .venv and gracefully prompt the user if GEMINI_API_KEY is missing in the .env file.
Actual behavior
1.Gemini login silently fails in the background with bash: gemini: command not found (Exit code 127).
2.The agent automatically converts hyphens to underscores, attempting to call inno_pipeline_planner, which results in a Tool not found error.
3.Upon tool failure, the agent falls back to run_shell_command and attempts to directly execute the skill folder itself, causing a No such file or directory or Is a directory crash.
4.Claude Code CLI hangs indefinitely at the Paste code here if prompted > prompt without launching a browser or showing an auth URL.
5.The gemini-deep-research script crashes first due to ModuleNotFoundError: No module named 'requests', and after manual installation, crashes again because GEMINI_API_KEY is not passed to the shell execution context.
VibeLab version
VibeLab v0.1.5
Operating system
macOS 26.1
Install method
npm global
Logs, screenshots, and evidence
Impact and severity
No response
Additional information
No response
Summary
This issue captures a series of critical friction points encountered during the initial setup and execution of VibeLab's research pipeline. The problems span across CLI authentication hangs, incorrect agent tool routing (specifically hyphen vs. underscore mismatches), inappropriate execution of skill directories, and missing environment/script dependencies.
Steps to reproduce
1.Gemini CLI Init: Attempt to log into Gemini via the VibeLab UI on a fresh environment where @google/gemini-cli is not globally installed.
2.Tool Routing: In the VibeLab Chat interface, prompt the agent with: "Please use the inno-pipeline-planner skill to start a research pipeline."
3.Claude Auth: Run the claude command in the VS Code integrated terminal to authenticate Claude Code.
4.Script Execution: Prompt the agent to execute the gemini-deep-research skill via shell commands.
Expected behavior
The system should prompt the user to install the missing CLI tool or handle the missing dependency gracefully.
The agent should correctly map the hyphenated skill name (inno-pipeline-planner) and recognize it as a conversational skill directory, initiating the interactive workflow to generate research_brief.json and tasks.json.
The claude CLI should successfully open a browser window for OAuth or provide a visible fallback URL in the terminal.
Python-based skills (like deep_research.py) should have their dependencies (e.g., requests) automatically resolved in the .venv and gracefully prompt the user if GEMINI_API_KEY is missing in the .env file.
Actual behavior
1.Gemini login silently fails in the background with bash: gemini: command not found (Exit code 127).
2.The agent automatically converts hyphens to underscores, attempting to call inno_pipeline_planner, which results in a Tool not found error.
3.Upon tool failure, the agent falls back to run_shell_command and attempts to directly execute the skill folder itself, causing a No such file or directory or Is a directory crash.
4.Claude Code CLI hangs indefinitely at the Paste code here if prompted > prompt without launching a browser or showing an auth URL.
5.The gemini-deep-research script crashes first due to ModuleNotFoundError: No module named 'requests', and after manual installation, crashes again because GEMINI_API_KEY is not passed to the shell execution context.
VibeLab version
VibeLab v0.1.5
Operating system
macOS 26.1
Install method
npm global
Logs, screenshots, and evidence
Impact and severity
No response
Additional information
No response