codex
garrytan/gstack / codex
runs Codex for a second opinion on your code , which improves: spots flaws faster
Details
- judge’s score
- 8/10
- place
- outside the packs
- publisher
- garrytan
- status
- new
- pinned commit
- c580561 on GitHub (new tab)
- skill’s last commit
- Oct 19, 2026
- repo stars
- 137,030
- main type
- Building with Claude
Judgement
Exact commands, real pitfalls (--base incompatible with a prompt, sandbox forced to read-only, timeouts). Vague description ("three modes") and very long boilerplate. SKILL.md truncated at 24k out of 59k.
reservations
- SKILL.md truncated at 24000 characters out of 59335: the end of the workflow and the output sections are not assessed.
- sections/*.md appendices each truncated at 5000 characters; the rest of the content was not seen.
- A lot of generic boilerplate (preamble, telemetry, voice, AskUserQuestion) that weighs down the skill without direct relation to Codex.
- gstack-* scripts in bin/ and lib/outside-review-result.ts referenced but not provided; description not very precise about when to use it.
- Triggering
- 1/2
- Added value
- 2/2
- Precision
- 2/2
- Scope
- 2/2
- Quality
- 1/2
Alerts
4 alerts, by category
Execution
- high risk
Sends the diff, the plan and repository files to a third party (OpenAI Codex); the skill does not clearly specify what to exclude (secrets, .env).
Network
- high risk
Repository content, potentially untrusted, is injected into a Codex prompt: risk of indirect prompt injection, mitigated only by a text instruction.
Instruction
- medium risk
The GSTACK_CODEX_NO_SANDBOX=1 option gives Codex full write access; use with caution.
Static scan
- info risk
allowed-tools frontmatter
Bash - Read - Write - Glob - Grep - AskUserQuestion
Original description
OpenAI Codex CLI wrapper — three modes. (gstack)
added on Oct 19, 2026 · license MIT