Home / Engine / Agents and Council Architecture

Making Hermes Effective - diagnosis + plan (v1)

Updated Jun 19, 2026 · Affirmology_Hermes_Effectiveness_Plan_v1.md

Making Hermes Effective - diagnosis + plan (v1)

Triggered 2026-06-18: Sol used Atlas, asked Hermes a real, sophisticated question, and got "horrific non-results." This is the diagnosis (a self-analysis / probe of Hermes's gaps) and the plan to fix it.

The prompt that failed (Sol)

I want to access Jeff and Sol's wealth codes using Gene Keys and Human Design, and astrology as appropriate. Also full numerology can be utilized too. Jeffrey Alan Parker and Soledad Gabriela Ballard, full names for full numerology. We are looking for information to help us break through where we are blocked when it comes to money and prosperity coming in. Perhaps the timing of this current Venus and Jupiter transit and Pluto movements also help for this time of 2026. Doesn't have to be a joint reading, I'm looking for information about each of us individually, actionable paths to prosperity. Also include an energetic blueprint for resources, abundance, and impact.

This is exactly what the SYSTEM is built to do. Hermes could not, and here is why.

Self-analysis: why Hermes failed

Hermes today (affirmology-studio/api/main.py, /api/hermes) is a SINGLE Claude call with a STATIC text prompt. That prompt contains only: the brand rules, a HOW-ATLAS-WORKS blurb, the support scope, and a few BASIC placements for the four people (Sun, Moon, Rising, type, Life Path). That is all Hermes knows. It has: - NO way to compute or read a full chart. It cannot see Jeff's or Sol's Gene Keys spheres (including the Pearl/prosperity sequence), Human Design channels, full astrology, or name-based numerology, even when Sol pasted the full names and birth data. It has no tool to turn birth data into a chart. - NO connection to the corpus (the ~17k tradition records sitting in the cloud), so it cannot ground an interpretation in real Gene Keys / HD / astrology source material. - NO connection to the specialist oracles (Sophia, Athena, Prometheus, Pythagoras). They are prompt designs in agents/council.py, not something Hermes can call. So Hermes cannot "talk to the oracles." - NO transit/timing engine, so the Venus/Jupiter/Pluto 2026 request is unanswerable. - NO access to the project / per-person profile docs (the deep material on Jeff's hard drive). The "jpl-knowledge" share is referenced in the Atlas spec but is NOT wired to the cloud at all.

So when Sol asked for wealth codes, Hermes had nothing to draw on but the model's generic priors and five placements. It either gives bland generalities or risks making things up. That is the "horrific non-result."

IMPORTANT REFRAME: the gap is NOT the AI model. Hermes already calls Claude (a strong model; anthropic_key_set: true). Giving Hermes "access to AI models" is already done. The gap is DATA and TOOLS: Hermes is a smart brain locked in an empty room. The fix is to open the doors to the chart engine, the corpus, the oracles, and the docs.

The good news (most of this already exists in the cloud)

The Studio's engine, which Hermes is sitting right next to in the same service, already does almost all of it: - chart.compute_chart(BirthData) computes the FULL chart from birth data. - reports/render.py render_all_facts(chart) returns full text for astrology, Human Design, all three Gene Keys sequences (Activation, Venus, and PEARL = prosperity), and render_numerology_facts includes Expression, Soul Urge, and Personality (the NAME-based numbers) plus Life Path, Personal Year, Pinnacle, and Challenge. So "full numerology with full names" is already computable; it just needs the confirmed full legal name (Jeffrey Alan Parker, Soledad Gabriela Ballard), which Sol provided. - chart/vedic.py vedic_facts_text(chart) adds the sidereal layer. - The corpus (corpus.db, ~17k records) is already pulled into the cloud on boot and is queryable via the engine's db.search_records(). - The oracle reasoning lives in agents/council.py (Sophia, Apollo, etc.) and the character voice profiles.

Hermes is NOT wired to any of it. That wiring is the whole project.

The fix: make Hermes AGENTIC (tool use)

Turn /api/hermes from one static call into a tool-using agent. The Anthropic SDK + key are already there. Give Hermes tools it can call mid-conversation: 1. get_person_chart(person_id) and get_chart_facts(name, birth_date, birth_time, birth_location) -> runs the engine + render_all_facts and returns the FULL chart for a seeded person or for any birth data pasted into chat. (This alone answers most of Sol's prompt.) 2. search_corpus(query, tradition) -> grounds interpretations in the real tradition records. 3. consult_oracle(system, question, chart_facts) -> runs the specialist (Sophia/astrology, Athena/Gene Keys, Prometheus/HD, Pythagoras/numerology) with depth, so Hermes genuinely "talks to the oracles" and routes. 4. (later) get_transits(date) and search_docs(query). Hermes orchestrates: pulls the charts, consults the right oracles, grounds in the corpus, and writes the answer, permission-walled.

The plan (phased)

PHASE 1 - FULL CHART ACCESS (build first; biggest fix; all dependencies already in the cloud). Give Hermes the chart tools (get_person_chart, get_chart_facts). Now Hermes can read Jeff's and Sol's full Gene Keys (incl. Pearl/prosperity), HD, astrology, and full numerology (with confirmed full names). Sol's prompt becomes answerable for the individual prosperity reading. Also: confirm/store the four people's full LEGAL names so name-numerology is correct (Jeffrey Alan Parker, Soledad Gabriela Ballard, Colin's + Josh's legal names).

PHASE 2 - CORPUS GROUNDING. Add search_corpus. Hermes's interpretations stop being generic and start citing the tradition material we scraped, same grounding the script generator uses.

PHASE 3 - THE ORACLES AS REAL CONSULTANTS. Add consult_oracle so Hermes can hand a question to Sophia / Athena / Prometheus / Pythagoras and weave their specialist answers. This is "talk to the other oracles," and it is where the council comes alive in chat.

PHASE 4 - TRANSITS & TIMING. Build the Transit & Timing computation (the astrology engine can compute current planetary positions; compare to the natal chart) + ground in the transits corpus. Answers Sol's Venus/Jupiter/Pluto-2026 ask. Bigger build (the Transit agent was designed, not built).

PHASE 5 - THE KNOWLEDGE BRAIN ("full access like the project chat"). Get the deep material off Jeff's hard drive and into the cloud so Hermes can read it: sync the project docs + each person's full profile docs to R2 (or a bundled knowledge folder, or the jpl-knowledge GitHub repo), then add search_docs. This is what makes Hermes approach the depth of prompting inside this project workspace. New incoming client accounts get stored in R2 too (a cloud people store), so Jeff/Sol/Colin can pull any person Hermes is permitted to see.

PHASE 6 - PERMISSIONS AT DEPTH. As Hermes gets real data, enforce the tiers already designed: the team sees the team's charts + the shareable material; a future client sees only their own. Private business/equity stays Jeff-only.

Recommendation / sequence

Build PHASE 1 now. It is mostly wiring (the engine, the facts generator, and the birth data are already in the cloud service), and it directly fixes the exact failure Sol hit. Then Phase 2 + 3 quickly after (corpus + oracles) to make the readings grounded and multi-voiced. Phase 4 (transits) and Phase 5 (the docs brain) are the larger follow-ons that reach Jeff's full vision. On "GitHub repo access": the reliable path is syncing the docs into the cloud (R2 or a bundled knowledge set) rather than live GitHub calls; we can mirror the repo into R2 on a schedule.

BUILD STATUS (2026-06-18) - Phase 1 BUILT (uncommitted, needs a push + live test)

Hermes is now AGENTIC. New affirmology-studio/api/hermes_tools.py + a tool-use loop in /api/hermes. Tools live and wired: - get_person_chart(jeff|sol|colin|josh) and compute_chart_from_birth(name, date, city, time) -> FULL chart facts (astrology, Human Design, all 3 Gene Keys sequences incl. Pearl/prosperity, full numerology incl. name-based numbers, Vedic sidereal layer) via the engine that already runs in the cloud. - search_corpus(query, tradition) -> grounds in the ~17k tradition records (cloud corpus). - consult_oracle(sophia|athena|prometheus|pythagoras|parashara, question, chart_facts) -> a real specialist sub-reading. So Hermes can now PULL the charts, CONSULT the oracles, and weave a grounded, actionable answer. Sol's wealth-codes prompt is now answerable. Hermes's system prompt updated to USE the tools and never answer chart questions from memory. Cost is bounded (max 6 tool rounds). Cannot runtime-test in the sandbox (Mac-only engine venv); the real test is on deploy (tool-use needs sonnet, ANTHROPIC_API_KEY is present).

STILL PENDING (Phase 2, what we need to figure out)

1. FULL VEDIC: the engine's Vedic is only the partial sidereal layer (rashis, nakshatra), NOT full Parashara (dashas, divisional charts, yogas). So Parashara reads shallow today. Build the Vedic engine + collect/store Vedic for Jeff & Sol (engine work in affirmology-agent; coordinate with the corpus session).
2. TRANSITS & TIMING: build the Transit agent (the ephemeris can compute current Venus/Jupiter/Pluto positions; aspect them to the natal chart; ground in the transits corpus) so the "timing in 2026" half of Sol's ask is answerable. Add a get_transits tool.
3. NAME-NUMEROLOGY ACCURACY: store each person's confirmed full LEGAL name (Jeffrey Alan Parker, Soledad Gabriela Ballard, Colin's + Josh's) so the name-based numbers are right; never guess a name.
4. THE KNOWLEDGE BRAIN: sync the deep project + per-person profile docs from the hard drive to the cloud (R2, mirrored from a GitHub repo on a NIGHTLY CRON per Jeff) + a search_docs tool. This is the "full access like the project chat" bar.
5. THE AGENTIC COUNCIL: today Hermes consults oracles independently; Jeff wants the oracles to TALK TO EACH OTHER (Persephone, Chiron, the relationship lenses, Echo, the gates all interacting). A real multi-agent orchestration, bigger build. Also wire Chiron + Persephone into the generation council.py.
6. NEW CLIENT ACCOUNTS in R2 (a cloud people store beyond the 4 seeded) + the client permission tier so Hermes can serve them their own data only.
7. COST/PERF: the agentic loop makes several model calls per deep question; tune model choice per tool (cheap model for chart/corpus, strong for synthesis) and cache. HOW MUCH OF THE COUNCIL IS WIRED (Jeff asked): the SCRIPT-generation loop (Sophia synthesis + Orpheus draft + Apollo's 4 gates) is wired in council.py and runs on every Studio render. Chiron and Persephone are prompts, not yet wired. The 5 tradition oracles are now consultable from HERMES (this build) but do not yet converse with each other. The relationship engine (relationship.py) is built but not exposed in Studio/Hermes. The Transit agent is unbuilt.

One correction to an old note

The Agent Registry calls numerology "date-based, name-independent." That is stale: render_numerology_facts already outputs Expression / Soul Urge / Personality from the name. So full-name numerology is available now; the real discipline is only using it with a CONFIRMED full legal name (the accuracy rule already flagged), never a guessed/partial name.