phab MCP: AI sees some reviewhelper bot comments, some not

[jensstutte]

moz MCP get_phabricator_revision: identical bot author renders both readable and redacted

Summary
The same comment author (reviewhelper-bot, a known Mozilla review-infra account)
has its comment body returned in full on one revision but stripped and relabeled
[Untrusted User] on another — within the same bug stack and the same session.

Repro

• get_phabricator_revision(307915) → reviewhelper-bot's "Summary of Changes"
  general comment renders with author name and full content.
• get_phabricator_revision(307916) → reviewhelper-bot's comments (one general
  • two inline) all render as author [Untrusted User] with body replaced by
    "[Content from untrusted user removed for security]".
• Both revisions are in the same stack (bug 2049092, D307915–D307919).

Deterministic, not transient
Re-fetching D307916 returns byte-identical redacted output. Retrying / looping
the call does not change the result, so this is not a flaky author-resolution
timeout that a retry clears.

[suhaibmujahid]

This should be resolved by #6171. I will deploy it today, where comments from review helper will not be redacted anymore.

However, the underling discrepancy you are pointing to seems to be pointing to a bug where the comment from reviewbot seems to be used to truse previous comments, which is not the intended behaviour. Based on the current deployed version, both should have been redacted.