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

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.

Start by tracing the get_phabricator_revision entry point and compare how comments from reviewhelper-bot are handled for revisions D307915 and D307916. Done means the same known bot author and comment content are handled consistently across both revisions, with regression coverage for the reported reproducer.