critchie.fyiProject directory

THE WORKBENCH

Ideas, in progress.

A home for the things we’re building.
Open a project, see where it stands, and follow what comes next.

PROJECTSSTATUS & PROGRESS
Development paused · not production-ready

Hermes Forge

A reusable Hermes skill that turns complex goals into planned, model-routed work with independent review and evidence-based acceptance.

Open project : Hermes Forge
Estimated completion45%
45%

About 45% of the full intended scope, estimated by weighted workstreams—not test counts. Durable state, recovery foundations, workspace guards and real Kanban API contracts are verified. Complete task execution with VM-backed verification, live model routing, general repair, project integration and final release acceptance remain unfinished. Work is saved at a recoverable checkpoint for a later resume; the link opens a status page, not an operational app.

MVP deployed · private access

DNA Explorer

Ask freeform questions about your own DNA variants against live authoritative sources, with evidence scorecards.

Open project : DNA Explorer Open on home network : DNA Explorer (private LAN access)
Acceptance status
Deployed and live-verified on the private network; real-genome and Codex-link gates pending

The MVP is running privately on the home network. A live end-to-end question was verified through the deployed app: a synthetic test genome was imported, real ClinVar, PubMed, Ensembl and GWAS Catalog records were retrieved, and a persisted answer reported an observed variant with source-linked evidence and personal-risk fields correctly left uncalculated. Import, matching, research, validation, persistence, backup and restore are built and tested (133 automated tests). Remaining explicit gates: importing and asking about the owner's real genome file, the optional OpenAI Codex account link (needs owner device approval), and a routine dependency reinstall when the package-security scanner is healthy. Not a diagnostic test; no medication guidance.

Development beta deployed · owner access

MercuryAlpha

A private, research-assisted paper portfolio workspace with deterministic risk controls and no live-trading path.

Open project : MercuryAlpha
Acceptance status
Two-company valuation beta with refreshable dated EOD inputs and immutable saved analyses

The public development site at mercury.critchie.fyi runs reviewed beta 069203a. Apple and Microsoft issuer-derived cash-flow valuation hypotheses support separate editable assumptions, sensitivity and modeled value-gap comparison. Actual FMP end-of-day captures flow into new analyses without restarting the app, while saved reports retain their original price observations. The installed authenticated Analyze, Save, reload and restart journey passed with all nine previous reports preserved. Refresh is an explicit operator action, not an automatic or live-price feed. Default conditional valuations do not establish buying opportunities, expected returns or alpha; qualified GARP screening and broader evidence remain incomplete. HTTPS, authentication and isolation remain enabled; the previous full application/data are retained separately and real-money execution remains disabled.

Beta deployed · owner access

AutoTrader

A private stock-strategy research and paper-trading workspace; real-money trading remains disabled.

Open project : AutoTrader
Acceptance status
Runtime and saved five-stock research deployed; real-data Paper demo unfinished

Reviewed beta 5ec49af is running in matching web and engine containers behind authenticated HTTPS. Research now links to runtime observations for SPY, QQQ, AAPL, MSFT and NVDA and to the separate saved historical/dated-quote report. The owner-login journey through Runtime, saved research, Strategies, Settings and Paper status was verified. Current retained runtime evidence reports 25 recipe slots not evaluated, not 25 losing trades or qualified opportunities; source checks remain Unknown and no current shortlist is admitted. Owner data and credentials were preserved with a verified matched recovery archive. Real-data Paper execution remains unfinished: Paper is paused with zero fills, and real-money trading is disabled/unarmed. This interim research release does not complete the required Paper demo or establish a continuous executable feed or profitability.

MVP deployed · owner access

News Curator

Scheduled, source-grounded news briefs with configurable topics, trend analysis, a live headline ticker and Telegram delivery.

Open project : News Curator
Acceptance status
Source collection and scheduling deployed; owner OAuth activation pending

The reviewed application is deployed behind the authenticated NPM profile with persistent schedules, history, settings and a dedicated OpenAI Codex OAuth home. Google News is live from the deployment network; GDELT was reachable but rate-limited during verification. Telegram delivery was accepted by the configured bot. The remaining owner action is to link ChatGPT / Codex from Settings and run the no-private-data activation probe before AI curation is claimed live.

Practical v1 core online · sign-in required

Resume Genie

Private résumé tailoring, cover letters and application tracking.

Open project : Resume Genie
Acceptance status
ChatGPT / Codex linking live; AI generation pending

Reviewed facts, versioned résumés, cover letters, exports and validated restore are implemented and tested. ChatGPT / Codex account linking is live under AI connection; complete your own fresh authorization on OpenAI’s official page. Linking does not activate AI generation: tool-free inference verification and an owner-authorized live-model test remain pending. No automatic applications, invented qualifications or guessed completion percentage.

Development paused · workspace online

AI Novel Studio

An idea-to-book writing studio, built around your creative direction.

Open project : AI Novel Studio
Estimated completion20%
20%

Roughly 20% of the full design. Manual editing, version history, basic exports and Codex sign-in are built. AI drafting is not yet available; continuity, reviews, research, images and publishing exports remain unfinished.

MVP deployed · private LAN

Dungeon Adventure

A text-first party adventure where an AI dungeon master narrates and fair server-side dice resolve every risky action.

Open project : Dungeon Adventure Open on home network : Dungeon Adventure (private LAN access)
Acceptance status
Deployed and live-verified on the private network; public hostname pending

The MVP is running privately on the home network. A full live end-to-end journey was verified through the deployed container: account creation, a solo-hero campaign, and a real GLM 5.3 dungeon-master turn that returned original narration, committed a validated objective, and honored duplicate-submit safety. Hidden d20-style checks, advancement points, party roles across three original worlds, and admin model/effort controls are built and tested (34 offline tests plus live checks). Images and audio are honestly shown unavailable because Ollama Cloud has no generation endpoints (verified 2026-09-24); text play is fully functional. Remaining: the dungeon.critchie.fyi hostname and HTTPS via Nginx Proxy Manager, which needs the owner's NPM admin login.

Privately deployed · final browser acceptance pending

Ghost Signal

Five-level cybersecurity investigation lab — 50 challenges, Linux, forensics, web security and incident response

Open project : Ghost Signal
Acceptance status
50 challenges and 10 bonuses runtime-tested; authenticated HTTPS deployed; final owner/browser gate pending

All five levels, ten bonus questions, persistent Mission Control, a Linux desktop, scoring, hints, casebook and backup/restore are implemented. The host suite passed 698 tests with one documented host-only skip; packaged challenge workflows and an independent private-promotion audit passed. The deployed route uses trusted TLS and the existing authenticated access profile; anonymous access is denied. Final browser navigation and desktop verification through the authenticated edge remain pending, as do human beginner and accessibility playtests. Vulnerable targets and raw VNC are not published.

MVP deployed · authenticated access

API Explorer

A visual API workspace with endpoint maps, editable requests, structured responses and Ollama-assisted documentation mapping.

Open project : API Explorer
Acceptance status
Reviewed source and deployed Ollama discovery/mapping verified; owner sign-in and FMP account testing pending

Deployed behind the existing authenticated NPM profile with trusted HTTPS. Saved connections, guided Ollama/FMP catalogs, OpenAPI JSON imports, manual endpoints, response trees/tables and redacted history/curl exports are implemented. The candidate passes 55 automated tests and independent review. The deployed application discovered 17 live Ollama models, executed a real catalog request and generated a GLM 5.3 endpoint proposal from fictional documentation. Anonymous public access is denied. FMP requests depend on the supplied key and subscription; live FMP and owner sign-in/browser acceptance are not yet verified. Arbitrary API keys do not reveal complete endpoint catalogs; supply a base URL and documentation/spec. Private-network APIs and OAuth flows are outside this version.

Dashboard deployed · authenticated access

OpenAI Usage & Cost Analytics

Date-range token, cache, reasoning and model analytics with actual API costs and model-cost comparisons.

Open project : OpenAI Usage & Cost Analytics Open on home network : OpenAI Usage & Cost Analytics (private LAN access)
Acceptance status
Hermes token telemetry and sourced API-equivalent cost scenarios live; organization API awaiting Admin key

The LAN dashboard is live and records Hermes OpenAI/Codex counters every 15 minutes, with model, provider, billing mode, input, cached input, output and reasoning views. It compares the observed token mix against OpenAI's dated Standard short-context API rates without double-counting reasoning tokens. Exact range deltas begin with the collector baseline; older data remains lifetime-only. Actual organization API usage and invoice costs are implemented but remain explicitly unavailable until an OpenAI organization Admin API key is stored through the named-secret workflow. ChatGPT web/app token detail is not exposed by OpenAI and is not fabricated.