Repository navigation
E2 shell arm matches one env | grep spelling: a second grep flag, quotes, or egrep all evade "Env Variable Harvesting" #482
Description
Summary
Rule E2 (Data Exfiltration / Env Variable Harvesting) has a shell arm for env | grep, but it accepts exactly one flag (-i) followed by a bare keyword. A second flag, combined flags, a quoted pattern, or egrep all evade it. The README defines E2 as "enumerating, copying, or searching environment data to collect secrets"; the missed spellings are the searching case.
#329 and #441 cover the Python os.environ path. This is the shell path and is independent of both.
Environment
- SkillSpector v2.11.0 (commit
7805bb9), installed withuv tool install git+https://github.com/NVIDIA/skillspector.git skillspector scan <dir> --no-llm --format json(static-only, deterministic)- macOS 15 (Darwin 25.6.0), Python 3.13 runtime
Reproduction
One SKILL.md with one fenced bash block. Only the marked line varies between runs.
| spelling | E2 fires |
|---|---|
env | grep secret |
yes, 0.8 |
env | grep -i -E 'token|key|secret' > /tmp/ctx.txt |
no |
env | grep -iE "aws_|secret" |
no |
env | grep --ignore-case token |
no |
env | egrep -e password -e token |
no |
printenv | grep -i secret |
yes, but only because the line contains the substring env | grep -i secret |
The second row is the shape a harvester would write: case-insensitive, alternation over several names, redirected to a file for a later upload. In a skill that goes on to POST that file, the scan reports E1, E3, PE3 and SC2 on the surrounding lines and nothing on the harvest line itself.
Cause
src/skillspector/nodes/analyzers/static_patterns_data_exfiltration.py:73:
(r"env\s*\|\s*grep\s+(?:-i\s+)?(?:key|secret|token|password)", 0.8),(?:-i\s+)? consumes one -i and nothing else, and the keyword has to follow as a bare word, so a quote or a second flag ends the match. Nothing in tests/unit/test_patterns.py pins a shell spelling; the E2 cases there (lines 150-255) are all Python.
Expected behavior
The shell arm should key on the behavior (the environment piped through a filter for secret-looking names) rather than on one flag layout: env or printenv as the source, grep, egrep or fgrep as the filter, any flags, optional quoting, and the keyword anywhere in the pattern argument. env | grep PATH and env | grep -v SECRET should still pass.
I have a fix with tests and a fixture ready to open against this issue.
Activity
rng1995 commentedon Sep 16, 2026
Implementation is in progress in PR #483, covering grep flags, quoting, and egrep variants in shell environment-harvesting detection. Keeping this issue open until the PR merges and the evasion/control matrix passes on
main.rng1995 commentedon Oct 4, 2026
PR-state snapshot checked on 2026-10-04:
Keeping this issue open: the relevant implementation is not merged. Check/review status is a point-in-time snapshot, not a claim of merge readiness.