codex-stack-selection

Inventory Codex skills, plugins, apps, and MCP servers and recommend project-scoped configuration changes.

Updated Jul 20, 2026
One-click install
npx skills add https://github.com/merceralex397-collab/alex-stack --skill codex-stack-selection-merceralex397-collab
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: codex-stack-selection
Source: https://github.com/merceralex397-collab/alex-stack/tree/main/dev/my-skills/general-skills/codex-stack-selection
Command: npx skills add https://github.com/merceralex397-collab/alex-stack --skill codex-stack-selection-merceralex397-collab

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Projects accumulate Codex skills, plugins, apps/connectors, and MCP servers over time, and it is hard to know which capabilities are actually used, which are failing, and which can be safely disabled without touching global configuration. ## Core Features & Use Cases - Stack Inventory: Runs a bundled Python script that enumerates installed skills, plugins, apps, and MCP servers using the Codex CLI, plugin manifests, and config files. - Usage Analysis: Scans local Codex session logs from the previous 14 days to attribute real project-scoped usage, counting failed attempts as use while redacting prompts, arguments, and credentials. - Evidence-Based Recommendations: Classifies each capability as Keep, Disable, or Needs attention using repository evidence and a documented selection policy. - Safe Configuration: Previews an exact .codex/config.toml delta and applies it only after explicit approval, validating the TOML afterward. - Use Case: Audit a repository's Codex setup, discover that an enabled MCP server has been failing for two weeks, and disable three unused plugins project-locally without changing global state. ## Quick Start Ask the assistant to inventory this project's Codex stack, analyze the last 14 days of usage, and recommend which skills, plugins, apps, and MCP servers to keep or disable.

Frequently Asked Questions about codex-stack-selection

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit which Codex skills and MCP servers a project actually uses?▼

Run the bundled codex_stack_inventory.py script against the project to list installed skills, plugins, apps, and MCP servers, then scan local session logs from the previous 14 days for project-scoped invocations. The report classifies each capability as Keep, Disable, or Needs attention.

How do I disable a Codex plugin or MCP server for one project only?▼

Add an enabled = false entry under the appropriate [plugins], [mcp_servers], [apps], or [[skills.config]] table in the project's .codex/config.toml. Project config is only honored for trusted projects, and the skill previews the exact delta before applying it.

Does the session log analysis expose prompts or credentials?▼

No. The scanner reads only the minimum transcript fields needed for attribution and never emits message text, tool arguments, results, tokens, environment values, or headers. Its optional cache stores only normalized capability identifiers, timestamps, statuses, and byte offsets.

What happens if no Codex session logs exist for a project?▼

Missing logs are reported as unavailable history, not as zero usage. Recommendations then rely on repository evidence such as AGENTS.md instructions, manifests, CI, and source entry points, and items may be marked Needs attention instead of Disable.

Can this skill uninstall global plugins or change global Codex config?▼

No. It only edits the project-scoped .codex/config.toml after explicit approval, and it never changes global config, uninstalls plugins, deletes skills, revokes authorizations, or adds telemetry hooks.

Why is a failing MCP server not recommended for disabling?▼

A failed call still counts as use, so the capability is classified as Needs attention rather than unused. The recommendation is to repair or replace it, and an auth_status of unsupported is treated as an inconclusive probe, not an authentication failure.