`docs/trust-boundary/boundary.md` § *the per-agent socket dir* already
explains why write permission on a directory confers the right to unlink
its entries and why the sticky bit does not restrain that. This comment
said the same thing in the same words, which is the "restating facts all
over" case rather than two independent explanations.
Measured rather than eyeballed, since "these look similar" is not a
finding: 6-gram intersection between this file's comments and that doc
went 36 -> 14. What remains is short shared phrasing — the
socket-activation sentence, which already carries its own pointer, and
the principals table, which is call-site material.
Kept at the call site, because a reader editing this line needs it and a
pointer alone would not stop the change:
* that 0751 is deliberate and security-relevant, with the consequence
named in one line;
* that `d` re-applies on every sync, so an external chown does not
survive the next agent's spawn.
That second one used to be written as history — a tug-of-war with a
`ChownSocketDir` unit that no longer exists. The constraint it carried is
still live, so it is stated in the present tense instead of as a story
about a removed mechanism.
Comments only; no behaviour change. Refs #3901.
0ec8a7d2f9