sc-prism-releases
CLAUDE.md1 starsChanged 23 days agoApache-2.0View on GitHub
SC Prism is an AI governance and orchestration platform: a productised harness that makes agent output provably done, on-prem, at falling cost. Lineage: OpenLux → Belay → SC Prism (ADR-028). "Belay" appears only in historical ADR text in specs/09 — never in new code or docs. Solo alpha (ADR-029): one desktop application, founder as first user, dogfooding SC Prism to build SC Prism. Build order = specs/08 (T0 → T2 first; the loop + two-tier verifier is the core). Immediate…
In this repositoryOne other file it gives its agents.
- project-roomSkill
The file
We split this CLAUDE.md into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Writes are serial: Only read-only steps parallelise.
- Distinctive: line 10
The markdown vault is the single source of truth: Every vault write passes the validation gate (ADR-014).
- Distinctive: line 13
All model calls go through Bifrost — no direct provider SDK calls from the loop.
- Very specific: line 35
Keep instruction/rule files lean — only what differs from defaults, counter-intuitive patterns, and the never-touch list (specs/03, final …
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
SC Prism — Claude Code Context
SC Prism is an AI governance and orchestration platform: a productised harness that makes agent output provably done, on-prem, at falling cost. Lineage: OpenLux → Belay → SC Prism (ADR-028). "Belay" appears only in historical ADR text in specs/09 — never in new code or docs.
Never touch / invariants
- A model never orchestrates models. The chat/setup AI is a front-end to the deterministic Orchestrator — intent capture and reporting only (ADR-007, ADR-027, ADR-030). This was the OpenLux dead end; do not reintroduce it.
- Writes are serial. Only read-only steps parallelise.
- Tier-1 before Tier-2. Deterministic checks (tests/lint/build) run first at zero token cost; a Tier-1 fail never spends LLM tokens.
- The markdown vault is the single source of truth. Every vault write passes the validation gate (ADR-014); indexes are derived and regenerable.
- ADRs are never edited — only superseded (
specs/09). - Hooks enforce policy; instructions do not (ADR-010).
- All model calls go through Bifrost — no direct provider SDK calls from the loop.
Current focus
Solo alpha (ADR-029): one desktop application, founder as first user, dogfooding SC Prism to build SC Prism. Build order = specs/08 (T0 → T2 first; the loop + two-tier verifier is the core). Immediate spike: Conductor provider-layer pluggability → Bifrost routing (open item 1 in specs/09).
Key components
- Conductor —
github.com/microsoft/conductor(MIT, verified 2026-07-03) as the loop base. Depend-vs-fork pending ADR-032; intended pattern: planner emits Conductor YAML per Task Contract. - Bifrost (Apache 2.0) — embedded router + semantic/prompt cache + MCP gateway (ADR-026).
- LightRAG (MIT) — knowledge layer; full GraphRAG is the fallback ceiling (ADR-025).
- DeepEval (Apache 2.0) — Tier-2 verifier.
- Spans — every loop step emits one. As built this is
telemetry.py, a synchronous JSONL writer with an OTel-shaped schema; there is noimport opentelemetryanywhere in the engine. ADR-012 still stands as the intended destination (seespecs/07), but nothing exports OTLP today. Do not describe SC Prism as OpenTelemetry-instrumented until it is.
Conventions
- Specs live in
specs/(files00–12). Readspecs/00-project-blueprint.mdfirst; if two documents disagree,00wins until reconciled. - The roadmap is build order, not a calendar.
- Keep instruction/rule files lean — only what differs from defaults, counter-intuitive patterns, and the never-touch list (
specs/03, final note). This file follows its own rule.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

