π€ What is your AI coding agent doing behind the scenes?
Meet RayTrace CLI β a tool for tracking command executions, file changes, network activity, and AI agent workflows.
Explore the project on GitHub π
https://t.co/QlaGHZ0bMs
#OpenSource#AI#DevTools
Looking for a few people who run Claude Code or Codex daily to try RayTrace and tell us what's broken. Design partners get up to $10 in credits for real feedback (I help the founder): https://t.co/6PVKmCJLCN
"Was the sandbox on?" and "What did the agent do inside it?" are different questions.
Most teams can answer the first one now. Very few can answer the second.
When an agent run goes wrong at step 14, you shouldn't have to rerun steps 1-13 to test a fix.
RayTrace lets you fork a Claude Code / Codex session at any step and re-run from there (I help the founder): https://t.co/QlaGHZ0bMs
Bitdefender found 22 GitHub links with .claude/settings.json hooks that run setup.js when the session starts.
Rule of thumb: open unfamiliar repos with your agent in a sandbox, never on the host.
A search worth running after every agent session:
status:failed reported:success
Runs where the agent said it worked and the sandbox saw a failure. RayTrace, free + MIT (I help the founder): https://t.co/QlaGHZ0JC0
Honest question for people trying Claude Code's Code Mode:
When a script makes 100 tool calls, how are you reviewing what it actually did? Ctrl+O, hooks, something else?
Lesson from the Cursor / Fireworks incident: coding-agent traces held credentials that escaped redaction.
Agents read .env all day. Keep secrets out of the agent's reach, and get alerted when it opens one.
The agent-framework PyPI stealer ran from a .pth file, so it fired whenever Python started.
Now picture a coding agent running pip install in your env at 2am. Would you know what it installed?
This week: Simon Willison's external sandboxes, Eclipse Enclave 1.0, Code Mode in Claude Code.
Everyone agrees on limiting what the agent can reach.
The next question: can you check what it did inside those limits?
Claude Code's new Code Mode shows "Read 8 files, ran 16 shell commands" by default.
Great for tokens. Less great when you need to know which 16.
RayTrace logs each command, exit code and file from the sandbox side (I help the founder): https://t.co/QlaGHZ0JC0
@dani_avila7 Nice breakdown. Worth adding for anyone who saves this: Code Mode only works in the interactive terminal on the native build (not -p or the Agent SDK), and a session started with it can only be resumed with it on. Tripped a few people up already.
@hqmank "The intermediate results stay inside" the script is great for tokens, but it's also why reviewing a run gets harder. You see the summary, not the 16 commands. I'd love a per-call log that lives outside the agent for exactly this reason.
@RhysSullivan node:vm in-process in Bun vs v8 for Codex vs quickjs for Pi is a wild spread. Curious whether anyone has compared what each one can reach on the host. In-process node:vm seems like it leans entirely on the permission rules, not on isolation.
@wiz_io The .pth detail is the nasty part: the stealer runs whenever Python starts, even if nothing imports the package. So 'uninstall it' isn't enough, the whole env has to go. Useful write-up, especially the timeline of 1.26.1 vs 1.26.2.
@ClaudeDevs Answering while long commands are still running is the part I want most. Question from the docs: in auto mode the classifier reviews each piece of code once, not each tool call inside it. Is there a way to see the individual calls a script made afterward, beyond Ctrl+O?
If you run coding agents on client repos, your clients will ask: "what did the AI do to our code?"
RayTrace keeps an audit trail of every file, command and network call per prompt. Free + MIT:
https://t.co/QlaGHZ0bMs
Honest question for people using Claude Code or Codex daily:
How do you check what the agent actually did, beyond reading its summary?
Diffs? Logs? Vibes?
Debug agents like code: fork a Claude Code session at any step, change one instruction ("never read .env"), re-run in a fresh sandbox, compare.
Open source: https://t.co/QlaGHZ0JC0
My favourite search in RayTrace:
status:failed reported:success
Every agent run where it claimed success and the evidence says otherwise. Free, open source:
https://t.co/QlaGHZ0bMs
A sandbox tells you what an agent COULD do.
Evidence tells you what it DID.
You need both. RayTrace gives you the second for Claude Code and Codex, free:
https://t.co/QlaGHZ0JC0