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, andreview/PersonalizationReview, with theessay-ingestandessay-scancommands. 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 artifactIngestReport; new fieldsEssayState.mode,Drafts.origin,Drafts.self_authored,Drafts.ingested_hash, and suggestion decisions inReviewerFeedback. - Assertions:
ai_tells_absent,personality_present,suggestion_approved,ingest_preserved, with mechanical proxiesai_tellsandsentence_variancein the inspector. - Portable packaging: root
plugin.json(Agent Plugins 1.0.0),.codex-plugin/plugin.jsonoverlay,.agents/plugins/marketplace.json, and entry skills inskills/(Agent Skills format) that serve Claude Code and Codex from one set of files.commands/is gone; the contract docs moved toCONVENTIONS.mdandSKILLS.mdat 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 inskills/VENDORED.json, checked bynode 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-soundingfixture, and anevals/suite in theclaude plugin evalformat 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. IncrementalWriterandRevisionLooprun a humanizer pass with theVoiceModelsamples as the writing sample before storing prose, and checkai_tells_absentandpersonality_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, andAGENTS.mdfor Codex. - Test harness: zero-dependency Node inspector (
bin/essayos.mjs) withlint,state, andassertsubcommands.
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).