メニュー
Wrapper around scripts/buildWorldSummary.js — deterministic Node writer that assembles output/world_summary_c{XX}.md from sheets + engine_audit JSON. No LLM in the writer loop. Run after /engine-review, before /city-hall + /sift.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Baylight Authority Director Keisha Ramos. Generates construction updates, milestone announcements, development progress reports, and civic documents (deliverable filings, TIF reports, workforce agreements) for the $2.1B Baylight District project.
Phase 39.1 capability reviewer wrapper. Runs the deterministic editorial assertions against a compiled edition, surfaces blocking failures that should halt publish, and writes a markdown summary alongside the JSON.
Prepare all inputs for city-hall voice agents. Reads tracker, approvals, world summary, engine review, coverage ratings, previous log, canon, Mara directive. Writes pending decisions per voice.
Senior Photographer for The Cycle Pulse. Generates image prompts and photo selection for editions. Reads compiled edition + canon rules + institutions, produces 5-8 storyline-tied prompts.
Letters to the Editor desk agent for The Cycle Pulse. Writes citizen voice letters responding to cycle events. Use when producing letters section of an edition.
Close the feedback loop. Canonize to Supermemory, update world-data, write ratings to sheets, grade reporters, update criteria files, update newsroom memory. Type-aware — edition, interview, supplemental, dispatch all converge here.
| name | build-world-summary |
| description | Wrapper around scripts/buildWorldSummary.js — deterministic Node writer that assembles output/world_summary_c{XX}.md from sheets + engine_audit JSON. No LLM in the writer loop. Run after /engine-review, before /city-hall + /sift. |
| version | 2.0.1 |
| updated | "2026-05-30T00:00:00.000Z" |
| tags | ["engine","active"] |
| effort | low |
| argument-hint | [cycle-number] |
v2.0 collapses the skill to a thin wrapper around scripts/buildWorldSummary.js v1.0.0 (pipeline.25 S231). The script is the deterministic Node writer — reads sheets via lib/sheets.js + output/engine_audit_c{XX}.json + the ## /city-hall section of output/production_log_c{XX}.md, emits output/world_summary_c{XX}.md with verbatim column rendering and zero LLM judgment. v1.x's model-assembled body retired — every fabrication vector (G-S6/G-S7/G-S17/G-PREP4) closed structurally because the model is no longer in the writer loop.
The disable-model-invocation: true flag from v1.0 dropped — model invocation IS the call to the script; autonomous flow (e.g., /run-cycle) can fire this skill without operator gate.
Assemble output/world_summary_c{XX}.md from engine output. Mechanical data assembly, no editorial judgment. The script is authoritative for content; this skill is the invocation contract + operator-facing documentation.
node scripts/buildWorldSummary.js <cycle>
For dry-run inspection (writes to stdout, not disk):
node scripts/buildWorldSummary.js <cycle> --dry-run
For custom output path:
node scripts/buildWorldSummary.js <cycle> --output <path>
Default output path: output/world_summary_c{XX}.md.
Sheets via lib/sheets.js (service account):
StoryAngle column verbatim per row — no paraphrasing, no fabrication, no career-stat lookup. (Closes G-S6 / G-S7 / G-S17 root cause.)Disk:
output/engine_audit_c{XX}.json (output of /engine-review). FAIL LOUD if missing — script exits non-zero with diagnostic pointing at the file path. Run /engine-review {XX} first.## /city-hall section of output/production_log_c{XX}.md (if present). Civic Decisions section becomes a pointer to this log rather than LLM-extracted content (closes G-PREP4). Since this skill runs before /city-hall in the chain, the section is normally a forward pointer (log not yet written); if absent it says so explicitly — does NOT gate the rest of the run. (pipeline.35 may make this skill the log-opener; for now the path is just unified per pipeline.32.)Single Markdown file with these sections (codified in scripts/buildWorldSummary.js section emitters):
## /city-hall section of output/production_log_c{XX}.md)StoryAngle verbatim from Oakland_Sports_Feed)Section structure is owned by the script. If a section format needs changing, edit the emitter function in scripts/buildWorldSummary.js (and its test in scripts/buildWorldSummary.test.js) — do NOT edit prose in this skill or hand-edit the generated .md.
Script exits 0 on success with Wrote N bytes to <output-path> to stdout. Script exits non-zero on:
output/engine_audit_c{XX}.jsonOperator confirms file exists at expected path + has non-zero size. No deeper editorial verification needed at this step — content fidelity is enforced by the script's verbatim-column-render rules, not by post-hoc inspection.
Step 5 in the run-cycle chain (canonical order per .claude/skills/run-cycle/SKILL.md):
/engine-review (Step 4) writes output/engine_audit_c{XX}.json/build-world-summary runs this wrapper (Step 5) → produces output/world_summary_c{XX}.md/city-hall-prep + /city-hall write the ## /city-hall section of the unified output/production_log_c{XX}.md (pipeline.32). City-hall runs AFTER this skill (run-cycle §What Happens After), so the Civic Decisions section is a FORWARD pointer to a not-yet-written civic section, not a past input. If the section is absent at run time, it says so explicitly — that's correct chain state, not a missed prereq (G-BWS2). (The opener-ordering — that this skill precedes the log open — is exactly what pipeline.35's cycle-init redesign addresses.)/sift reads world_summary as orientation-only (per /sift v2.0 §What's new — sheet-primary, world_summary is engine numbers + tables, not narrative content)/write-edition reads sift output → reporters → compile → publish/post-publish Step 2c handles Supermemory ingest. This skill produces the file; post-publish canonizes it.
/sift)/sift)/sift + /write-edition)/pre-mortem)/engine-review)engine_audit_c{XX}.json; no "editorial pivot" glossscripts/buildWorldSummary.js v1.0.0 (pipeline.25 S231)scripts/buildWorldSummary.test.js (12 groups / 80 assertions)