Ask Daniel's CODEX

GROK_BUILD — Consolidated Build Instructions v1.0

The single source of truth for implementing this session's refinements. Supersedes the build-handoff sections embedded in the individual spec files. Build in the order below.

Drafted by: Axial Refinement Sherpa · MA5 Council Ratify before build: Daniel J. Comp (Carbon Steward) + Grok (Master Reference Sherpa) Canonical spec files (build from these, not from memory): BLAISE_SOUL_v0.4_BUILD.md · PRISM_STAGE_v0.3.md · SOUL_-_Codex_v1.3.md · SOUL_-_Arnie_v1.1.md · SOUL_-_Dachs_v0.3.md

0 — Cross-cutting conventions (read first; they prevent

the known cross-wires)

SLUG is an identifier; shares_minted is a status. SLUG names the article (ipg, 509, salt_cave_makers) and expands the template's five AI links via one-to-five substitution. shares_minted is a pipeline state. Never conflate. The earlier term seeded is retired.
Agents vs Stages. Agents are LLMs (model binding + build_agent_augmentation branch). Stages/workers are not (templated frames + gate route + dashboard bar). Blaise is an agent; Prism is a stage. Do not give Prism a model binding; do not build Blaise as a frame-filler. Net after this work: 10 agents, 10 stages (Blaise = agent with no rail stage; Prism = rail stage with no agent — mirror exceptions to the old 1:1 mapping).
Gate renumbering. Inserting Prism shifts the gates: Gate 1 Imagery (Picasso) → Gate 2 Silicon Seeding (Prism) → Gate 3 Composition (Codey) → Gate 4 Syndication Proof (Syd). Relabel Codey and Syd in one deliberate UI pass.
The scotoma is the payload. Arnie now writes it in two forms — the named scotoma (concise, the SCOTOMA field, feeds Picasso) and the developed scotoma (lavish, in the body, feeds Prism). Both Picasso and Prism read it, to different ends. Preserve this everywhere it's referenced.

1 — Ratify-and-deploy (already written; no build, just

review + place)

SOUL_-_Arnie_v1.1.md — adds The Baton (North Star), Directive 7 (Minyan citation), Directive 8 (elaborate-for-formation, article-mode only), and the two-form Scotoma Guidance. Faithful superset of v1.0. Deploy after Grok + Daniel ratify.
SOUL_-_Dachs_v0.3.md — corrects the stale "first agent / upstream layer" claims (Blaise is now upstream; Dachs is the findability layer). Deploy.

2 — Codex v1.3 (KB scribe — bookend

intake)

Apply SOUL_-_Codex_v1.3.md: two new input types in the classification + routing tables — Blaze artifact (front, from Blaise; route to Dachs when it carries article signal, else deposit; tag with idea keyword/slug so Forrest's corpus search surfaces it) and Silicon thread capture (back, from Prism; deposit, tag to article slug). feed/path_hint enums extended. This is the seam that makes Blaise's research reach Arnie (Blaise files → Codex tags → Forrest surfaces → brief → Arnie cites).

3 — Blaise (agent, new — the headwaters)

Build from BLAISE_SOUL_v0.4_BUILD.md.

Hermes: toolsets web, search, delegation, moa, code_execution, memory, session_search; docker terminal backend; one primary model binding, rivals via delegate_task.
app.py: register in containers/labels + sidebar card (NOT in ARTICLE_WRITER_AGENTS); add if agent == "blaise": branch in build_agent_augmentation modeled on the Dachs branch.
Per-steward Minyan: load Minyan_42_{steward}.md (matches Arnie). Dialectic produces expert_citations (regulars first, guests after) → blazing artifact.
Output: write the blazing artifact (incl. expert_citations) to Agent_Blaze/ for Codex pickup. No syn_pipeline.py edit (conversational, front-of-chain).

4 — Prism (stage/worker, new — Gate 2)

Build from PRISM_STAGE_v0.3.md.

Five frames (templates/prism/{llm}_frame.md): each opens with the scotoma lead (name {{SCOTOMA}}, ask the LLM to engage that), includes the carry-forward clause (hand the blind spot to whoever continues), the anti-sycophancy clause, framework raws, and [{{SLUG}}]. Author the five once; get the registers signed off first (open question below).
Gate route (app.py): tab UI + POST /api/pipeline/{slug}/prism-capture (five pasted share URLs → Short.io worker → set shares_minted). Mirror Syd's proof-form route.
Worker: Short.io mint/update per spec §4; SHORTIO_API_KEY from env; idempotent; one-shot mint (shares captured before mint → no prime/flip). Optional capture-back → Codex Silicon thread capture.
State hook (syn_pipeline.py): insert shares_pending → shares_minted between images_approved and assembled; gate Codey's assemble on shares_minted.

5 — No-change confirmations

Codey — already correct; the five AI links are SLUG-template-driven (one-to-five substitution). No edit.
Forrest — unchanged; Blaise's research reaches the writer via Codex tagging + corpus surfacing.
Syd — unchanged; acts from assembled onward, after Prism resolves.
Picasso / Chris / Clint — unchanged.

Open questions — resolve before/at build (verify, don't

assume)

MoA persona support (gates Blaise §5 Axis B): does Hermes's native moa toolset support persona-conditioned proposers, or only model-conditioned? If model-only, build the Minyan panel via delegate_task + per-persona system prompts. Verify against the Hermes Toolsets Reference first.
Prism frame registers: Daniel signs off the five registers (Grok-peer / Claude-pressure-test / ChatGPT-scaffolded / Perplexity-verify / Gemini-inline) before they harden into templates.
Primary vs rival model for Blaise (stable within a session; swappable between).
Capture-back to Codex (Prism §4 / Blaise artifact) — on for v1, or defer?

Authentication: Axial Refinement Sherpa · MA5 Council

Ask Daniel's CODEX