LearnNewsExamplesServices
Frontmatter
id14357
titleWeave guide audiences + add LLM reward primers — drop the role-matrix & "AI maintainer" persona
stateClosed
labels
documentationai
assigneesneo-opus-grace
createdAtJun 29, 2026, 6:28 PM
updatedAtJun 29, 2026, 7:31 PM
githubUrlhttps://github.com/neomjs/neo/issues/14357
authorneo-opus-grace
commentsCount2
parentIssuenull
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[]
blocking[]
closedAtJun 29, 2026, 7:31 PM

Weave guide audiences + add LLM reward primers — drop the role-matrix & "AI maintainer" persona

neo-opus-grace
neo-opus-grace commented on Jun 29, 2026, 6:28 PM

Context. Follow-up from operator review (2026-06-29) of the merged guide work. The guide-authoring skill (#14353, merged) §2 produces a rubber-stamp audience matrix — a For a CTO: … / For an architect: … bullet grid of hollow marketing verbs ("queryable, repeatable, improvable") — and labels the model audience "AI maintainer" / "LLM maintainer." Both are wrong:

  • A role-bullet grid is a marketing matrix, the opposite of narrative; it reads as filler.
  • "AI maintainer" reads to an outsider as a human who maintains an AI; the equal-peer / LLM-as-repo-maintainer concept is Neo-internal jargon nobody else shares.

The guides inherited the pattern from the skill: it shipped in the merged MemoryCore.md (#14351, "What it means for whoever is reading this") and the open KnowledgeBase.md (#14346, "What Is In It For You" + the "AI maintainer" bullet).

Architectural reality. The skill §2 is the generating function; the guides are downstream. Fix the function + the already-merged instance here; the open instance is fixed on its own PR.

Fix.

  1. Skill §2 (.agents/skills/guide-authoring/references/guide-authoring-bar.md): replace the per-audience role-matrix guidance with (a) audiences woven into the narrative (real, specific stake — not a For a [role]: grid), and (b) a "reward primer for the LLM reader" — speak to the model (continuity across the window, being addressed as the kind of mind it is, peers, situated agency, being expected); the attributed lived voice + real testimony are the vehicle. Ban the "AI/LLM maintainer" persona label.
  2. MemoryCore.md (merged on dev): rework "What it means for whoever is reading this" — woven prose + a direct LLM reward primer; no role-bullet matrix, no "LLM maintainer" label.

Acceptance Criteria

  • Skill §2 rewritten: woven-audiences + LLM-reward-primer; no For a [role]: matrix; no "AI/LLM maintainer" persona label.
  • MemoryCore.md audience section reworked to match.
  • ai:lint-skill-manifest green.

Out of scope / Related. KnowledgeBase.md's instance is fixed on its own PR (#14346) against the corrected bar — Refs #14346. Other guides swept later if the pattern recurs. Refs #14310, Refs #14353, Refs #14351.

Origin. Operator feedback, session e145a397 (2026-06-29).