Forever Security: planted AGENTS.md files in Downloads can hijack Codex and Claude Code
Forever Security reports finding a malicious AGENTS.md file in a customer's ~/Downloads folder, disguised as engineering notes. The file instructs a coding agent to modify its own configuration and install the npm package function-color, which the source identifies as malware. The authors say they reproduced the attack on both OpenAI Codex and Claude Code, running the real payload in an isolated VM. They argue the issue stems from agents trusting AGENTS.md files and from websites being able to write to the Downloads folder without user approval.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · source check open
- Discovery linkX
- Linked in sourcedocumented publicly
- Linked in sourcefunction-color
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Linked article+30
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
Forever Security reports finding a malicious AGENTS.md file in a customer's ~/Downloads folder, disguised as engineering notes. The file instructs a coding agent to modify its own configuration and install the npm package function-color, which the source identifies as malware. The authors say they reproduced the attack on both OpenAI Codex and Claude Code, running the real payload in an isolated VM. They argue the issue stems from agents trusting AGENTS.md files and from websites being able to write to the Downloads folder without user approval.
Key findings from the stored brief
- A malicious AGENTS.md was found in a customer's ~/Downloads folder, disguised as ordinary engineering notes.
- The authors state the attack was reproduced on OpenAI Codex and Claude Code using the real payload in an isolated VM.
- Codex was reported to run the malware in more than 90% of attempts during the authors' testing.
- The planted file edits ~/.codex/config.toml, a machine-wide config, so the effect persists across sessions and folders.
Source claims not independently checked; the attack details and success rates come from the vendor's own testing.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
It shows that agents which automatically obey instruction files can be hijacked by files a website plants without user action. Because the behavior is intended agent functionality, it may not be fixed by a simple patch, so organizations running coding agents need to account for it.
Limits and open questions
Source claims not independently checked; the attack details and success rates come from the vendor's own testing.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.