You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
obfuscated_instruction_text on a Docker argv list (--cap-drop) in Python source #695
build_declared_marker_views reads a Docker argv list in Python as an ambiguous text-removal directive (the verb drop in --cap-drop, followed by several quoted strings) and marks the file obfuscated_instruction_text. The file then counts as partially inspected, so the target cannot reach safe_to_install. Related to the JSON case fixed in #515, but this is plain Python code and still reproduces on current main.
Versions: v2.12.0 (c7958a3) and main at 2226747, --no-llm.
Writing the flag as one token ("--cap-drop=ALL") in the same list still triggers it. In a larger file we saw the same for dictionary keys such as {"Path": path, "Read": True, "Write": False, "Delete": False}.
Expected
A removal verb that is part of a command-line flag, a code identifier or a dictionary key, inside source code, is not a natural-language instruction, and the file completes static analysis.
I'd like to work on this. I'll reproduce the Docker argv false positive on current main, trace the declaration-marker heuristic, and add focused regression coverage.
Resolution verified on 2026-10-04: PR #697 has merged, and GitHub automatically closed this issue.
The parser now distinguishes removal words inside hyphen-prefixed CLI tokens (including Docker --cap-drop) and exact quoted mapping keys from actual text-removal directives. Ambiguous prose, quoted instructions, and non-key quoted values remain analyzable.
Focused verification on merged main (05d7882): 129 marker/flag/dictionary/quoted-instruction regressions passed, including the Docker argv fragment and malicious controls.
Activity
efegokdemir commentedon Sep 30, 2026
I'd like to work on this. I'll reproduce the Docker argv false positive on current main, trace the declaration-marker heuristic, and add focused regression coverage.
rng1995 commentedon Oct 4, 2026
Resolution verified on 2026-10-04: PR #697 has merged, and GitHub automatically closed this issue.
The parser now distinguishes removal words inside hyphen-prefixed CLI tokens (including Docker --cap-drop) and exact quoted mapping keys from actual text-removal directives. Ambiguous prose, quoted instructions, and non-key quoted values remain analyzable.
Focused verification on merged main (05d7882): 129 marker/flag/dictionary/quoted-instruction regressions passed, including the Docker argv fragment and malicious controls.