Skip to content

dhruvaldarji/essayos

v1.1.0MIT

An idempotent, stateful, agentic operating system for writing application essays. It interviews you, drafts one section at a time in your voice, reviews an essay you already wrote, and checks that every claim traces to a real experience.

Changelog

All notable changes to EssayOS are documented here. The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

[1.1.0] - 2026-09-19

Added

  • Ingest flow: system/Ingest, architecture/ReverseOutline, review/AITellScan, and review/PersonalizationReview, with the essay-ingest and essay-scan commands. An essay the applicant already wrote is stored verbatim, cut into sections, claim-mapped, grounded by a targeted interview, scanned for AI tells and monotone prose, and revised one applicant-approved suggestion at a time under the ratchet. New artifact IngestReport; new fields EssayState.mode, Drafts.origin, Drafts.self_authored, Drafts.ingested_hash, and suggestion decisions in ReviewerFeedback.
  • Assertions: ai_tells_absent, personality_present, suggestion_approved, ingest_preserved, with mechanical proxies ai_tells and sentence_variance in the inspector.
  • Portable packaging: root plugin.json (Agent Plugins 1.0.0), .codex-plugin/plugin.json overlay, .agents/plugins/marketplace.json, and entry skills in skills/ (Agent Skills format) that serve Claude Code and Codex from one set of files. commands/ is gone; the contract docs moved to CONVENTIONS.md and SKILLS.md at the root.
  • Third-party skills: humanizer 3.0.0 and simple-english 2.1.0 bundled as pinned copies in skills/ (loaded by Claude Code and Codex alike), recorded in skills/VENDORED.json, checked by node bin/essayos.mjs skills-sync, and also listed in the marketplace as optional standalone installs pinned to the same commit.
  • Spec and evals: ISA criteria ISC-123 onward, linter checks for manifests, vendored versions, eval-suite shape, and plain-English docs, a tests/fixtures/ai-sounding fixture, and an evals/ suite in the claude plugin eval format with a credential-gated CI job.

Changed

  • Interview questions and applicant-facing reports follow the simple-english Plain rules (CONVENTIONS.md §5a, §9). The essay prose is exempt by contract.
  • IncrementalWriter and RevisionLoop run a humanizer pass with the VoiceModel samples as the writing sample before storing prose, and check ai_tells_absent and personality_present.
  • README, AGENTS.md, CONTRIBUTING.md, and the command text were rewritten in plain English.

[1.0.0] - 2026-06-23

Added

  • Initial release of EssayOS, an idempotent, stateful, agentic operating system for writing application essays (college admissions, scholarships, graduate school statements, fellowships, grants, and more).
  • Kernel: Orchestrator (scheduler + convergence), AssertionEngine (named quality + system assertions), LearningLayer (revision history and lessons).
  • Skills: 39 skills across discovery, architecture, writing, review, verification, and meta categories, all obeying one universal contract (CONVENTIONS.md).
  • Specialists: 6 reviewer-persona agents, also registered as Claude-plugin agents.
  • Schemas + templates: 18 machine-readable artifact schemas and matching blank templates.
  • Plugin packaging: .claude-plugin/plugin.json, slash commands, and AGENTS.md for Codex.
  • Test harness: zero-dependency Node inspector (bin/essayos.mjs) with lint, state, and assert subcommands.

Design

  • Core invariants: quality-aware idempotent artifacts (upsert-by-id, staleness propagation, per-input epoch) and a monotonic best-draft ratchet (median-of-k denoise + ε-margin + ceiling gate + terminal cap).