Hub rules-codex

rules-codex

v5 public Verified

Codex

Community-governed security ruleset for Codex

@sigmashakeinc 0 pulls 12 rules published Apr 10, 2026 synced Oct 11, 2026 sigmashakeinc/rules/rulesets/rules-codex →
Severity
4 error 4 warn 4 info
demo.cast
Community-governed security ruleset for Codex

Rules index

12 governance rules

Indexed from the repository's .rules files, grouped by technology. Expand any rule to read the raw source.

codex

audit-self-healing info log

Codex 'Self-healing' mode activated. It will autonomously run tests and iterate on fixes.

Rule source requires an active Pro or Enterprise subscription.
audit-self-healing info log

Codex 'Self-healing' mode activated. It will autonomously run tests and iterate on fixes. Ensure iteration limits and manual review checkpoints are configured to prevent runaway loops or silent breakage.

Rule source requires an active Pro or Enterprise subscription.
block-self-healing-force warning ask

Blocked 'Self-healing' attempt with forced dependency installation.

Rule source requires an active Pro or Enterprise subscription.
block-self-healing-force warning ask

Blocked 'Self-healing' attempt with forced dependency installation. Force-installing dependencies bypasses lockfile integrity and can introduce supply chain risks. Review why the dependency resolution failed before allowing a force install.

Rule source requires an active Pro or Enterprise subscription.
codex-no-auto-exec-generated-code error block

Generated code is being evaluated or executed directly. Codex completions are untrusted strings — executing them without sandboxing and review enables arbitrary code execution. Display completions for user review, run in a locked-down sandbox, or apply a strict allowlist of safe AST patterns before execution.

Rule source requires an active Pro or Enterprise subscription.
codex-no-prompt-injection-in-input error block

Prompt injection pattern detected in Codex API input. Embedding instruction overrides in user-controlled text allows attackers to hijack the model's behaviour, bypass safety layers, and generate malicious code (CWE-77 prompt injection). Sanitize or refuse user input before passing it to the Codex API. Never interpolate raw user content into the system prompt.

Rule source requires an active Pro or Enterprise subscription.
codex-no-secret-in-prompt error block

Hardcoded secret passed to Codex prompt. Embedding real keys/passwords in prompts causes them to appear verbatim in completions, logs, and billing dashboards. Load credentials from environment variables or a secrets manager and never interpolate them into completion requests.

Rule source requires an active Pro or Enterprise subscription.
codex-no-shell-true-generated error block

Generated Python code uses subprocess with shell=True and references a Codex completion. Passing model output to shell=True enables command injection — any shell metacharacter in the completion becomes an injection vector. Use shell=False with an argv list, or reject completions containing shell metacharacters.

Rule source requires an active Pro or Enterprise subscription.
limit-self-healing-iterations warning log

Codex 'Self-healing' has exceeded the threshold of iterations. This could indicate an infinite loop.

Rule source requires an active Pro or Enterprise subscription.
limit-self-healing-iterations warning log

Codex 'Self-healing' has exceeded the threshold of iterations. This could indicate an infinite loop or a fundamentally broken state. Stop the repair loop, investigate root cause, and restart from a known good state.

Rule source requires an active Pro or Enterprise subscription.
warn-autonomous-pr info log

Codex is creating an autonomous Pull Request.

Rule source requires an active Pro or Enterprise subscription.
warn-autonomous-pr info log

Codex is creating an autonomous Pull Request. Review the diff for unintended changes — Codex may silently alter unrelated files during self-healing or repair tasks.

Rule source requires an active Pro or Enterprise subscription.

Version metadata

Published metadata

The hash-locked record of this published ruleset. When source contents change, the maintainer resyncs and the version bumps.

Version
v5
Published
Apr 10, 2026
Source commits
0
Synced
Oct 11, 2026
Hash
3110adb9ef68c35e854f0afa0cec1d32fe53af7b625f13ffa3be2362c899b46c
Signature
Verified (publisher key)
Tested with
claude-code@current
Plan
Starter (free)
Visibility
Public

Source history

Recent commits

Latest commits touching rulesets/rules-codex/.sigmashake/rules in sigmashakeinc/rules/rulesets/rules-codex. The published version only bumps when a maintainer resyncs.

No commit history available.