Trojan Source and bidi overrides

Direction controls reorder how text is displayed without touching how it is parsed. In source, that makes the program a reviewer reads and the program the compiler builds two different things.

U+202E RIGHT-TO-LEFT OVERRIDEU+2066CVE-2021-42574bidi attack

The comment that is not a comment

$ augur scan auth.go
auth.go (text): 5 finding(s)

BIDI
   [alarm] offset 73 — U+202E RIGHT-TO-LEFT OVERRIDE
   [notice] offset 76 — U+2066 LEFT-TO-RIGHT ISOLATE
   [notice] offset 100 — U+2066 LEFT-TO-RIGHT ISOLATE
   [notice] offset 103 — U+2069 POP DIRECTIONAL ISOLATE

WHITESPACE
   [notice] offset 99 — 1 trailing whitespace character(s)
       1 bytes

The override sits inside a comment. Displayed, the line reads as though the privileged branch is guarded; parsed, the guard is not there and the function returns on the other branch.

It is not only source files

The same characters in a hook command, an MCP server entry or a permission allowlist mean an agent executes one thing while the file appears to say another. augur scans those files too, and reports the finding as a JSON path rather than quoting the line, because those files hold tokens.

Removed, and then checked from disk

$ augur clean auth.go
wrote auth.clean.go
verified: 5 removed, 0 finding(s) deliberately left in place

$ augur scan auth.clean.go
auth.clean.go (text): nothing hidden found

Removable, unlike a lookalike letter: deleting a direction control does not change what any word says. It is still a decision — a file that genuinely mixes scripts may want them — so the viewer lets you take them one at a time.

$curl -fsSL https://raw.githubusercontent.com/dejo1307/augur/main/install.sh | sh