docs(security): Codex independent pre-release audit prompt (anti-anchoring)#30
Merged
Conversation
…verlay - 2026-research-grounded prompt for an INDEPENDENT second-opinion audit by OpenAI Codex CLI (gpt-5.1-codex-max), anti-anchoring by design (RED attack register only, GREEN/PURPLE writeups deliberately withheld) - AGENTS.audit.md persistent execution rules (read-only sandbox, no network, no commits, 90 min, evidence-bundle 4-tuple confidence) - 12 adversarial recon targets (T1 audit_chain per-invocation, T2 azure transport-error tenant leak, T4 stop-hook summary unredacted, T5 YAML merge-key/anchor inert-filter gap, T7 serde_yml parse-time unsoundness)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
2026-research-grounded prompt + AGENTS overlay for an independent Codex CLI audit of CLX v0.8.1 / [Unreleased]. Anti-anchoring design (RED attacks only; GREEN/PURPLE writeups withheld; arxiv 2603.18740 anchoring effect cuts vuln detection 16-93%). 12 ranked recon targets including T1 (audit_chain.rs per-invocation, not chain), T2 (azure transport tenant leak), T4 (stop-hook summary unredacted), T5 (YAML merge-key inert filter gap), T7 (serde_yml parse-time before filter). Evidence-bundle 4-tuple confidence (no bare percentages). Read-only sandbox + no network + no commits + 90 min RoE. Ready to run via codex-rescue subagent or local Codex CLI; invocation example included.