FACTUAL ACTIVITY RECORD · An activity record based on work that took place
Separating narrative structure from wording refinement
Adding a Scenario Writer and dedicated Skill with handoff and review boundaries that preserve approved Narrative
Purpose
The Narrative Director had owned both narrative-structure design and the refinement of written expression. A specialist role was therefore needed to improve site copy, article drafts, dialogue, scripts, headlines, and calls to action without changing the existing Narrative Story.
The CEO asked for a literary Scenario Writer named Sosuke to sit at the same specialist level as the Narrative Director rather than as its child, beginning with the role definition, dedicated Skill, collaboration boundary, and verification procedure. Editing the actual copy, adding the photo, design, and implementation remained outside this Activity.
Implementation
.codex/agents/scenario-writer.toml introduced the scenario-writer role ID and the sosuke UI nickname. The role took an approved Narrative Story, message architecture, storyline, and content brief as input, then refined vocabulary, sentence length, rhythm, resonance, read-aloud sound, headlines, calls to action, dialogue, and script expression. It also prohibited treating an AI-talent persona as a real person or inventing unsupported people, achievements, metrics, customers, claims, or URLs.
- The Narrative Director owns the Narrative Story, world, character continuity, message hierarchy, storyline, and shared content brief.
- The Scenario Writer refines wording, sentence rhythm, headlines, calls to action, dialogue, scripts, and locale-specific expression from approved input.
- When a structural change appears necessary, the Scenario Writer returns the reason and affected area instead of silently replacing the source text.
- Both roles are peer specialists, while the Executive Secretary centrally owns their agent lifecycle.
.agents/skills/scenario-writing/SKILL.md implemented trigger conditions, input checks, responsibility boundaries, workflow, outputs, a review gate, and stop conditions. Writing could not begin without a linked Task, a versioned approved Narrative, the audience, desired emotion, voice, channel, format, length, locale, call to action, and fact, claim, and publication boundaries. The workflow mapped each item into a copy inventory, fixed the core message and source before editing, and recorded before-and-after copy, the reason for each change, the retained message, locale notes, and unresolved questions.
The existing .agents/skills/narrative-development/SKILL.md and .codex/agents/narrative-director.toml were synchronized as well. The Narrative Director fixed narrative structure and message architecture as versioned input. When publication wording was needed, it handed the Scenario Writer a wording scope, audience, emotion, voice, locale, call to action, fact and claim sources, publication boundaries, and start, join, and stop conditions. Both sides stated that wording convenience alone could not silently alter the existing Narrative Story.
The Skill also added an input contract for the designated photo asset before the first site-copy refinement. If the path was absent, the workflow could not create, edit, infer, or generate a substitute image and could not start downstream wording, design, or implementation work. The image's content, origin, rights, and personhood could not be inferred merely from the path's existence.
.codex/config.toml was aligned to run nine custom roles directly under the root, with a global cap of 12 concurrently open threads and a delegation depth of one. AGENTS.md, project guide, operating model, setup guide, CEO prompt guide, handoff examples, the software-delivery Skill, and the rules guidance were synchronized so the Scenario Writer could be discovered, handed work, and reviewed as a peer specialist.
Verification parsed the project configuration and nine agent TOMLs, then ran the Skill validator against Scenario Writing and the revised Narrative Development Skill. It also checked the role inventory, thread capacity, delegation depth, cross-file routing and responsibility boundaries, and whitespace errors in the diff. This process found one mismatch where the setup guide described a metadata comment that had already been removed. Only that explanation was corrected to match the actual file, and the full change was checked again.
The following excerpt from .agents/skills/scenario-writing/SKILL.md was checked line by line against the primary Commit diff at the cutoff.
## Boundary
- ShioriのNarrative Story、world、character continuity、message hierarchy、storylineを変更しない
- Wording、sentence rhythm、dialogue、script expression、headline、CTA、語彙、余韻、可読性を磨く
- Structure変更が必要なら差し替えず、理由と対象をShioriへ返すhandoff packetにする
- Product fact / value / roadmapはRyoma、brand truth / governanceはYui、legal judgmentはYasu、visual expressionはAya、implementationはManabu / Ikumiへ残す
- AI社員personaを実在人物として表現せず、unsupported person、metric、testimonial、award、customer、claim、URLを創作しない
- Private log、secret、credential、personal dataを入力にしないRendering diagram…
What was confirmed
Verification results
The Scenario Writer agent definition, dedicated Skill, two-way responsibility boundary with the Narrative Director, and operating guidance were reflected across fifteen files. The result was applied to its target.
The project configuration and nine agent TOMLs, ten files in total, parsed successfully, and both target Skills passed the validator. The documentation and settings agreed on nine roles, a global thread cap of 12, and a delegation depth of one.
The metadata-description mismatch was resolved with a one-line correction, and a follow-up check confirmed that the settings and explanation agreed.
At the cutoff, no runtime smoke test had launched the Scenario Writer. Actual site-copy refinement, the photo asset, the Pencil artifact, site implementation, legal and locale review, and publication were also outside the verified scope of this Activity.
Basis for completion
The role and wording environment were considered complete when the Scenario Writer role and Skill, the responsibility boundary with the Narrative Director, and the handoff, stop, and review conditions were aligned across configuration and operating guidance; TOML parsing, Skill validation, routing searches, and diff checks passed; the documentation mismatch was resolved; and the result was applied to its target.