ethos-rig

Agentic harness for one-shotting the Ethosism book corpus in /Users/charlie/Ethosism/ethos.

The harness scans the corpus, scores print readiness, creates a task queue, renders per-agent prompts, and can optionally dispatch Codex tasks against the target repo.

Quick Start

cd /Users/charlie/Ethosism/ethos-rig
npm run validate -- --target /Users/charlie/Ethosism/ethos
npm run one-shot -- --target /Users/charlie/Ethosism/ethos

This writes generated state to /Users/charlie/Ethosism/ethos/.ethos-rig/ by default:

• inventory.json: discovered books, chapters, word counts, support files, slop signals, print-gate evidence, and target-owned publication/handoff evidence.
• plan.json: readiness scores, score dimensions, publication status, and task contracts.
• tasks.jsonl: durable task queue.
• prompts/: one prompt per bounded agent task.
• reports/one-shot-plan.md: human-readable plan.

When the target already contains a current, clean publishing/output/print-gate/corpus-print-gate.json, validation can legitimately produce zero manuscript tasks and zero prompts. The CLI still prints the print-gate status, print-order packet counts, launch handoff counts for session evidence receipts, operator checklist steps, transfer files, and bundle files, account handoff counts for title-creation evidence cards, commercial handoff counts for evidence rows and review cards, proofing handoff counts for evidence cards and review checklists, retail handoff counts for contact tracking, order intake, order confirmation, review copies, event planning, event sales reconciliation, consignment terms, shelf talkers, library acquisition requests, public quote permissions, returns/damage claims, vendor onboarding, sample/excerpt requests, media inquiries, bulk/adoption inquiries, errata/correction rows, accessibility/accommodation rows, contact permission rows, boundary/referral rows, photo/recording permission rows, inventory receiving rows, inventory allocation rows, delivery confirmation rows, invoice templates, payment remittance rows, sell-through/reorder rows, packing slips, shipment logs, and support files, reader-use counts for paths, path selection worksheets, session invitation templates, session follow-up templates, session materials checklists, session plans, session reading-cycle rows, session agenda cards, session question cards, session evidence review cards, session role cards, session participation agreements, session repair cards, facilitator role agreements, participant notices, session attendance trackers, session accessibility support rows, session boundary/referral rows, participant handouts, practice commitment trackers, session closeout summaries, participant feedback cards, facilitator intake rows, facilitator feedback rows, and title cards, external publication blocker count, next launch action, launch policy, handoff freshness, handoff launch-blocker rollup rows, print-only entrypoint audit counts, current-title intake smoke counts, and handoff archive paths with SHA-256s so 0 manuscript tasks is not mistaken for publication completion or for missing transfer artifacts. The plan report also shows soft readiness deductions such as word-mass or prose-cleanliness scores, surfaces external publication blockers from publishing/output/status/, publishing/output/launch/, and publishing/output/handoff/, checks whether the handoff commit, blocker rollup, and archive hashes still match the target, and summarizes the active launch waves, next-session workbook start row, immediate external batches in next-session order, launch session receipt, operator-checklist, transfer, and bundle counts, account title-creation card counts, commercial review-card counts, proof evidence-card counts, print-order transfer counts, retail handoff and reader-use counts, and final upload/source/launch-session/current-title/print-only/smoke checksums from the target launch CSVs and handoff manifest.

To dispatch agents:

node ./bin/ethos-rig.mjs one-shot \
  --target /Users/charlie/Ethosism/ethos \
  --execute \
  --concurrency 1

Use concurrency carefully. Book-writing tasks share corpus state and should usually run one at a time unless the books are clearly independent.

Architecture

The layout follows the Agent-Skills harness pattern:

• iso/: source of truth for cross-runtime instructions, command router, agent roles, and MCP config.
• modes/: progressively loaded operating procedures.
• templates/: executable policy, book system, scoring, contracts, and states.
• lib/: deterministic corpus inventory, scoring, planning, prompt rendering, and Codex dispatch.
• bin/: user-facing CLIs.
• batch/: durable worker prompt and runner wrapper.
• docs/: workflow and pipeline references.

Commands

node ./bin/ethos-rig.mjs inventory --target /Users/charlie/Ethosism/ethos
node ./bin/ethos-rig.mjs plan --target /Users/charlie/Ethosism/ethos
node ./bin/ethos-rig.mjs prompts --target /Users/charlie/Ethosism/ethos
node ./bin/ethos-rig.mjs one-shot --target /Users/charlie/Ethosism/ethos
node ./bin/ethos-rig.mjs validate --target /Users/charlie/Ethosism/ethos

Design Lineage

From Agent-Skills, this harness keeps the split between deterministic package behavior and target-owned corpus state. From autonovel, it adapts the layer stack and modify-evaluate-keep loop: corpus canon, book frame, manuscript chapters, print/export support, and review state.