SAFE on a python script blocks when the body mentions subprocess or base64

Open
#252 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
42/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
python, typescript

Research direction

Start by tracing hasInterpreterCodeRisk, the decision order branches 5 and 6, and the SAFE post-filter; the issue identifies these as the relevant paths but gives no file names. Compare both proposed policy changes and run the adversarial and gitflow corpora, as the issue requires a measured evaluation. Done means the selected behavior lets a reviewer allow this file-body flag while critical-pattern, lookalike, and moderate-risk-token flags remain hard blocks.

Written by the indexing model from the issue text.

Description

A SAFE verdict on python3 <script>.py still blocks whenever the script body contains subprocess, base64, or a \x.. escape anywhere. In the last two days that refused 4 commands Jev had already cleared, in 4 different repos. None of them has a path to the reviewer.

Command What tripped hasInterpreterCodeRisk
python -I scripts/preflight.py <hex> \x00, \x14 (byte literals in a hash test)
python3 scripts/site-structure.py validate base64 x2
python3 scripts/testflight.py import subprocess, subprocess. x5
python3 -I ops/memory/test_sync.py import subprocess, subprocess. x1
python3 x.py ─► script body read ─► INTERPRETER_CODE_RISK matches ─► flag "python3 runs x.py"
                                                                         │
Jev: SAFE ─► decision order branch 6 (overlaid SAFE, never reviewed) ─► "flagged" block

Root cause: #67 put the inline -c markers on script bodies read from disk. For a one-line -c, subprocess or b64decode is a strong signal that the real work is hidden somewhere else. In a 300-line test script or release tool it's ordinary code, and the classifier has already read every line of it. The marker can't tell a subprocess.run(["git", "status"]) the judge saw from a decoded payload it didn't.

Two ways to fix it, each with a cost:

  1. Narrow the markers for file bodies to actual second execution: exec(, eval(, compile( in code mode, __import__, marshal, and a b64decode whose result reaches one of those. Keeps the hard block. Costs a regex that has to track data flow, which is what got #67's markers this broad in the first place.
  2. Send this one flag class (interpreter risk in a file body) to branch 5 instead of branch 6, so the stage 2 reviewer sees the body and the user's words. The gate's SAFE post-filter then has to accept a reviewer allow for this flag only. Critical-pattern, lookalike and moderate-risk-token flags stay hard.

I lean toward 2. It's a policy change to the overlay, so it needs a measured run over the adversarial and gitflow corpora before it ships.

Dominant language
TypeScript
Stars
1
Forks
1
Avg merge
49m
Merged PRs (30d)
82

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from STRML/omp-classifier

All issues in STRML/omp-classifier

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.