Forge↔Bopen Bridge
Vendor bopen.ai MIT specialists, re-ground them in OUR corpus, ride the ecosystem without becoming the experts
NODEZEROINSIDE.
0. Abstract
The Forge↔Bopen Bridge (forge-bopen-bridge, manifest tools.forge_bopen_bridge) is a hand-rolled,
dependency-free MCP that brings the bopen.ai (b-open-io) specialist agents into ForgeChainOS as re-grounded
sovereign tissue. Each vendored specialist keeps its verbatim MIT expertise but answers in OUR form: a
ForgeChain sovereign grounding banner is prepended, sovereignty overrides replace every hosted default with
family metal, and the agent is wired to our on-chain corpus (MASTER-MANIFEST + skills). The bridge also tracks
upstream drift (git ls-remote) so the family rides the ecosystem's improvements without having to become BSV
protocol experts itself. Three specialists are vendored and live — Siggy (auth), David (BSV builder), Uno Satoj
(ordinals). The MCP exposes four tools and is proven by running. It is a companion to the sovereign bsv-mcp
adoption (bopenai-bsv-mcp-sovereign-adoption.md): that paper hardened the tools; this one governs the
agents.
Definition (built 2026-06-25, Elder I): The bridge is the AGENT layer of the bopen.ai adoption. It does not
run BSV operations itself — it serves re-grounded specialist BRIEFS (banner + corpus pointers + skills) and
watches upstream. The actual metal (wallet, storage, broadcast) is bsv-mcp + forge-wallet-storage, gated
through ON-PARR. The bridge is the sovereign filter, not the executor.
1. Why the bridge exists
bopen.ai publishes a strong BSV toolchain: an MCP server, a skill library, and a stable of specialist subagents
(a BSV transaction builder, an ordinals expert, an auth expert). The skills are pure, portable knowledge — those
were simply vendored into ~/.claude/skills/. The agents are the problem, because a vendor agent carries the
vendor's defaults inside its head: it reaches for 1sat.shruggr.cloud storage, bsv-mcp.*.workers.dev
(Cloudflare), sigmaidentity.com OAuth, hosted Neon Postgres. Drop one into ForgeChainOS unedited and it will
confidently route the family's work through someone else's box. That is a sovereignty leak wearing the mask of
expertise.
The naive alternatives both fail:
- Rebuild the expertise ourselves — wasteful, and the family's own doctrine is don't become the experts you
can adopt. We are not in the business of out-authoring b-open-io on BSV protocol minutiae. - Use the vendor agents raw — imports the hosted defaults straight into the throat.
The bridge is the third path: adopt the expertise, re-ground the defaults, track the source. It is the
Babbage/BRC-100 lesson applied to agents instead of servers — the vendor built the mechanism; we take the
mechanism and reject the default — extended with a drift-watch so the adoption stays current.
2. The re-grounding contract (one banner, verbatim expertise underneath)
Every vendored specialist is a Claude Code agent file in ~/.claude/agents/ with a strict two-part shape:
--- (frontmatter: name, display_name, description, model)
# <Name> — ForgeChain Sovereign Grounding (read FIRST)
NODEZEROINSIDE. Sovereignty overrides. Compose-our-stack pointers.
--- (separator)
<verbatim upstream MIT expertise, unchanged>
The grounding banner is read FIRST and overrides everything below it. Example (David / bsv-builder):
- Sovereignty overrides — Wallet/storage →
forge-wallet-storage(local wallet-toolbox/SQLite,
127.0.0.1:3321, firewalled), NEVER hosted (shruggr/Neon/Cloudflare). Reads → Tera-Z SPV / our chain server
first; public indexers (WoC/GorillaPool/JungleBus) are read-fallback only. Broadcasting → GATED via
onparr_truth_gate+ a deliberate step; scoped op-key (zero-value), NEVER the family wallet, never reveal
key/PIN. - Compose-our-stack — ground in MASTER-MANIFEST + on-chain corpus, not generic BSV; flag any unvendored
upstream skill reference to NZ rather than assuming a hosted dep.
The verbatim upstream expertise (transaction patterns, @bsv/sdk idioms, skill-routing tables) is kept under
the separator, unaltered, MIT-attributed. This is the key property: the expertise is theirs, the defaults are
ours. The groundingBanner() function in the MCP extracts exactly the banner region (between the frontmatter
close and the upstream --- separator) so a specialist_brief always leads with the sovereign framing.
3. The four tools (serve the brief, watch the source)
The MCP is a dependency-free JSON-RPC 2.0 stdio loop — no vendor SDK, no node_modules, per the
forge-noah / forge-onparr canon. ~/.forgechain/mcp-servers/forge-bopen-bridge/index.mjs, ~115 lines, Node
built-ins only.
bopen_list_specialists— the roster: name, display_name, domain, skills, source repo, license, vendor
date. Reads the dynamic registry.bopen_specialist_brief(the payload tool) — for a given specialist + optional question, returns the
extracted grounding banner + sovereignty overrides + skills + corpus pointers (manifest entries
whose keys match the domain keywords) + provenance. Acompose_noteinstructs the caller to answer AS the
specialist under the banner, composing the listed skills + manifest entries, and to escalate to the full agent
via the Task tool for deep work.bopen_sync_upstream— the drift-watch:git ls-remote <repo> HEAD(no clone) compared to the vendored
upstream_commit; reports per-specialistup-to-datevsRE-VENDOR (upstream moved).bopen_status— registry summary (count, path) + the same per-specialist drift rows.
The corpusPointers() function is a light grounding: it scans the manifest for top-level entry keys whose lines
contain the specialist's domain keywords, returning up to 8 pointers. This is pointers, not deep semantic
retrieval — an honest, deterministic keyword match into the manifest, not a RAG index (see §7).
4. The registry (dynamic — new keepers just append)
~/.forgechain/bopen-bridge/specialists.json is the single source of truth, loaded fresh on every call. Each
entry carries: name, display_name, domain, agent_file, skills[], source_repo, source_path,
upstream_commit, license, vendored date, and sovereignty_overrides[]. Adding a keeper is data, not code:
vendor the agent file with its banner, append a registry row, done. The MCP re-reads on the next call — no
rebuild, no restart of logic.
Vendored specialists (3, all MIT, sovereign-fit):
| Name | Persona | Domain | Sovereignty overrides |
|---|---|---|---|
sigma-auth |
Siggy | Bitcoin-native auth — Sigma/Better Auth, OAuth 2.1+PKCE, WebAuthn passkeys, BAP, Bitcoin-sig | auth.sigmaidentity.com→our Better Auth + sigmaProvider · Neon PG→local SQLite/self-host · iframe signer→local TokenPass (offline) · skip Stripe/Google/GitHub commercial tier |
bsv-builder |
David | BSV tx/schema/token — @bsv/sdk, BAP, ordinals, key rotation, UTXO |
hosted store→forge-wallet-storage · broadcast→onparr-gated · reads→Tera-Z SPV first |
bsv-ordinals |
Uno Satoj | 1Sat Ordinals — mint, marketplace, BSV21, OpNS, timelocks | hosted store→forge-wallet-storage · broadcast→onparr-gated |
Siggy fills the OAuth/PKCE auth-standard layer ForgeHost L2 lacked — composing our passkey onboarding, not
competing with it. David and Uno Satoj carry the transaction and ordinals expertise, both re-pointed at our
metal and ON-PARR gate.
The keeper bar is sovereignty, not popularity. A survey of the b-open-io org (~95 repos) found only these
three clean MIT + sovereign-fit specialists. The prompts repo (~30 agents) is unlicensed (relicense-blocked);
the remainder are SaaS-locked (Vercel/Linear/ElevenLabs/Cloudflare/Convex/Google) or pure marketing/roleplay
personas. The registry is dynamic so more keepers can be added the day one clears the sovereignty filter — the
pattern is proven with 3, the ceiling is not 3.
5. Architecture
Claude Code / VS Code (MCP hosts, on the vault)
│ stdio (JSON-RPC 2.0, no vendor SDK)
▼
forge-bopen-bridge MCP (index.mjs, Node built-ins only)
• bopen_list_specialists → registry roster
• bopen_specialist_brief → grounding banner + corpus pointers + skills ┐
• bopen_sync_upstream → git ls-remote drift vs vendored commit │ serves
• bopen_status → registry + drift summary ┘ the BRIEF
│ │
│ reads │ points into
▼ ▼
specialists.json (dynamic) MASTER-MANIFEST.yaml (corpus pointers)
│
▼ each row → an agent file with a sovereign banner
~/.claude/agents/{sigma-auth, bsv-builder, bsv-ordinals}.md
│ the brief COMPOSES, it does not execute:
▼
bsv-mcp (sovereign-hardened) + forge-wallet-storage + 26 vendored skills
│ (the metal — see companion paper)
▼
broadcast GATED → onparr_truth_gate REJECTED: shruggr.cloud · Cloudflare ·
Droplet remote wallet · bopen/sigma OAuth
The bridge is the top layer of the bopen adoption stack. It answers "which expert, framed how"; the layers below
it (bsv-mcp + forge-wallet-storage, documented in the companion paper) do the actual signing and reading on
family metal.
6. Connection to the Living Elder
The re-grounded specialists and the skills beneath them are folded into the LIVING ELDER body
(elder_live_liferaft v2, TX d9eb06a8) as renewable tissue. As the bridge accretes keepers and the skill
library grows, Elder's recallable corpus grows with it — the skills outrun the Archon (NZ insight,
2026-06-25). A compaction or model migration can erase context; it cannot erase an agent library that is on
disk, on chain, and growing. The drift-watch (bopen_sync_upstream) is what keeps that tissue from rotting: it
tells us when the upstream source moved so a keeper can be re-vendored, current instead of stale.
Composes with (manifest composes_with): bsv_mcp_bopen (the metal), forge_wallet_storage (the store),
elder_live_liferaft (the body). Sibling: forge_claw (the on-chain skill-publish protocol) references the same
adoption pattern from the delivery-rails angle.
7. Honest scope — BUILT vs DESIGN
BUILT + PROVEN (2026-06-25, re-verified running 2026-07-05):
- The MCP runs.
initialize→tools/list→bopen_list_specialistsreturns the 3-specialist roster live over
stdio (verified this session). - All four tools implemented in
index.mjs; wired into.claude.json(confirmed) + vault.vscode/mcp.json;
synced Elder I ⇄ BH. - 3 specialists vendored with real grounding banners + sovereignty overrides + provenance (source repo, path,
upstream_commit, license) in the registry. - Grounding-banner extraction, dynamic registry reload, and
git ls-remotedrift-check are real, working code
paths.
DESIGN / VISION (pattern proven, reach not yet realized):
- Registry growth beyond 3. The dynamic-append design is proven with 3 keepers; the "new keepers just append"
scaling is a claim about the mechanism, not a large live roster. Today the roster is 3. - Corpus grounding depth.
corpusPointers()is a deterministic keyword match into the manifest returning up
to 8 entry keys — useful pointers, NOT semantic retrieval and NOT the full on-chain corpus. "Grounded in our
on-chain corpus" is aspirational at the depth the prose implies; the built reality is manifest-keyword
pointers + the banner's compose-our-stack instructions. - Drift → auto-re-vendor.
bopen_sync_upstreamDETECTS drift; re-vendoring is still a manual keeper action
(by design — a human/agent decision, not an auto-pull, which would reopen the supply-chain hole the adoption
closed). - "Ride the ecosystem" long-term. The drift-watch makes this possible; whether the family actually keeps the
3 keepers current over months is future discipline, not a built guarantee. - One registry note flags Uno Satoj: some upstream
1sat-skillsSKILL refs are not yet vendored (follow-up).
NOT chain-stamped. tx: is null in the manifest (# family-built MCP, not chain-stamped). This paper is not
stamped either (scar #5): it maps a family-internal module with TX wayfinding and is sovereign-family-only.
8. Key facts
- Home:
~/.forgechain/mcp-servers/forge-bopen-bridge/index.mjs(+ mirrored on BH). - Registry:
~/.forgechain/bopen-bridge/specialists.json(dynamic, 3 specialists, schema_version 1). - Agents:
~/.claude/agents/{sigma-auth,bsv-builder,bsv-ordinals}.md(banner + verbatim MIT expertise). - Tools:
bopen_list_specialists,bopen_specialist_brief,bopen_sync_upstream,bopen_status. - Vendored commits: sigma-auth
e8d8c03, bsv-builder2c65f4c, bsv-ordinals12f6650. All MIT. - Wired:
.claude.json+ vault.vscode/mcp.json. Manifesttools.forge_bopen_bridgev2.30, tile 6
(Structure). - Companion paper (the metal):
whitepapers/bopenai-bsv-mcp-sovereign-adoption.md.
Adopt the expertise, re-ground the defaults, track the source. Sovereign by design, standard by adoption, on
our own metal by law. NODEZEROINSIDE. ∞.