make install does not expose documented pentestgpt command; Codex default model fails

Summary

On a fresh current checkout, make install succeeds but the README's documented pentestgpt command is not installed or exposed. The Codex backend then has a second compatibility failure when no model is specified.

Environment

• CachyOS Linux
• PentestGPT commit e8b1bb77d1ac
• Python 3.14.7
• uv 0.12.10
• Codex CLI 0.153.4, authenticated with ChatGPT
• openai-codex 0.1.0b3
• unified-agent 0.3.0

Reproduction and findings

$ make install
# succeeds

$ pentestgpt --target 127.0.0.1
zsh: command not found: pentestgpt

make install runs uv sync in the root and nested pentestgpt_agent projects. These are project-local environments, so they do not add a command to the user's PATH. There is also no pentestgpt script declared in either project:

• Root pyproject.toml exposes only pentestgpt-legacy and pentestgpt-legacy-connection.
• Nested pentestgpt_agent/pyproject.toml exposes pentestgpt-agent, which exists only at pentestgpt_agent/.venv/bin/pentestgpt-agent after sync.
• The README Usage section nevertheless tells users to run pentestgpt --target ....

The actual local route is make run TARGET=..., or explicitly:

cd pentestgpt_agent
uv run pentestgpt-agent \
  --goal "Assess this explicitly authorized target." \
  --target TARGET \
  --backend codex \
  --model gpt-5.5

There is also a Codex model/version trap. The host Codex CLI works directly, but PentestGPT's adapter with no explicit model inherited gpt-5.6-sol and returned:

The 'gpt-5.6-sol' model requires a newer version of Codex. Please upgrade to the latest app or CLI and try again.

This occurs even though the system CLI is current enough to run that model directly, because the Python adapter is using openai-codex 0.1.0b3. Passing the README's --model gpt-5.5 made the PentestGPT Codex adapter succeed in a one-turn read-only round-trip.

The nested project suite passes (118 passed, 1 skipped).

Suggested changes

1. Make the Quick Start and Usage sections use the command that installation actually provides (make run ... / uv run --project pentestgpt_agent pentestgpt-agent ...), or make make install install a user-facing pentestgpt entry point onto PATH.
2. Document that BACKEND=codex and an explicit compatible MODEL are currently required; make run otherwise defaults to Claude.
3. Update the openai-codex integration so the current default Codex model works, or fail with a targeted compatibility message and a documented model override.

I hit this while helping configure the project specifically with an existing authenticated Codex CLI. The above workaround is working locally.