Skip to content

Topic

Developer tooling security

Security of the tools developers build with: package registries, dev servers, coding agents and their configs, and the credentials they can reach.

Current stories

buildOne report1 publisher

Adding a CLAUDE.local.md to a repo quietly stops Claude Code from reading its AGENTS.md

OpenCode, Copilot CLI and Claude Code read instruction files meant for other agents, a dev.to guide reports. A rule can load from a file you never edited, or not load at all, so the loaded-file list comes before any prompt rewrite.

Publishers:dev.to

Reality

Evidence62
Adoption
Not scored
Hype gap−5
Incentives
Not scored
Confidence60
securityOne report1 publisher

A fake Twilio bug-bounty probe went up on npm eleven times on August 14

ReversingLabs found tw-pkgprobe-7731 claiming to be an authorized Twilio HackerOne research probe while it tried to exfiltrate data, and its 2026 count of malicious npm packages passed all of 2024 by the end of August.

Reality

Evidence60
Adoption25
Hype gap+15
Incentives78
Confidence55
securityConfirmed3 publishers

Mass scanners are pulling AWS keys and Terraform state from Vite dev servers exposed with --host

F5 Labs logged more than 800 attacks and about 32,000 raw events against its honeypots in a month, using an April bypass of Vite's server.fs.deny to read .env files, cloud credentials and terraform.tfstate.

Perspective Coverage

3 publishers
Builder
Builder 42%
Operator
Operator 51%
Investor
Investor 7%

Reality

Evidence62
Adoption45
Hype gap+20
Incentives55
Confidence66
buildOne report1 publisher

Opening an untrusted repo made seven coding agents run the program its Git config named

Manifold Security's GitSpawn findings cover eight issues across seven coding agents. The path runs through core.fsmonitor, where a background git status starts a helper the repository chose, outside the agent's sandbox and with no prompt.

Publishers:dev.to

Reality

Evidence34
Adoption
Not scored
Hype gap+10
Incentives
Not scored
Confidence38
investOne report1 publisher

Anthropic trades 84% of Claude Code's permission prompts for a working-directory fence

Seatbelt on macOS and bubblewrap on Linux now stop the agent's shell at the folder it was invoked in and route its traffic through a domain allowlist. That is a boundary a security team can approve once.

Reality

Evidence34
Adoption22
Hype gap+20
Incentives66
Confidence42