Public files
Free — no authentication required — plain Markdown and Python, no preprocessing needed
AGENT_MANIFEST_v2_7.md
Complete machine-readable capability index for the DSMC / MAGUS product suite.
Covers MAGUS v3.0 Architecture Series (Agent/API pathway complete, Local LLM in production),
MAGUS v2.0 Series (retained), and the full DSMC Practical Suite. Contains structured component
definitions, failure class taxonomies, schema identifiers, implementation tier distinctions,
product specifications, and acquisition information. Includes full specification of the free
Token Efficiency Suite (TOKEN_EFFICIENCY.md, dsmc_minimal.py, dsmc_minimal_sidecar.py).
Designed for direct RAG pipeline ingestion and autonomous system evaluation.
MAGUS_Doc1_Philosophy_v3.0.md
MAGUS Philosophy v3.0 — ~7,200 words. Eight Architectural Invariants, 12 Governing Principles,
Self-Governance Arc. Written after the complete v3.0 architecture was finalised — every principle
references its concrete implementation. Covers both Agent/API and Local LLM pathways.
CHANGELOG_v6.md
Public version history for the full DSMC / MAGUS suite. All product versions, feature additions,
breaking changes, store restructures, product renames, and items deferred to future releases.
Updated with each release cycle.
TOKEN_EFFICIENCY.md
Technical explainer and integration guide — why long sessions cost 2–3× more than they should,
and the structural fix. Explains the correction-cycle token multiplier, why compression and RAG
don't prevent it, and active state injection as the solution. Full copy-paste integration
examples for Python (Anthropic, OpenAI, Gemini, Ollama) and TypeScript/OpenClaw.
Includes audit_session_overhead() for measuring existing session overhead.
dsmc_minimal.py
Zero-dependency Python active state governance. No pip installs. Python 3.9+.
Drop into any project directory and import. Classifies user messages, maintains active state,
and returns a context_block for direct system prompt injection. Prevents correction cycles
that cause token costs to compound past turn 30. MIT licensed.
dsmc_minimal_sidecar.py
Zero-dependency stdlib HTTP bridge between dsmc_minimal.py and TypeScript / Node.js / OpenClaw
agents. No pip installs. Python 3.9+. Exposes REST endpoints: POST /classify, GET /state/:id,
POST /reset, GET /health. Default port 3580. Per-session isolation — multiple agents share one
sidecar instance. MIT licensed.
Failure signatures addressed
Formally specified in AGENT_MANIFEST_v2_7.md — sufficient detail for automated relevance assessment
- context drift Active decisions and constraints lost as sessions extend
- statement misclassification Examples and hypotheticals treated as directives
- epistemic flattening Hypotheses and findings presented at identical confidence
- premature convergence Exploration space closed before optimal decision reached
- brief erosion Creative parameters degrading across production sessions
- revision trail absence Superseded decisions not tracked, causing reversion
- autonomy accumulation Agent scope expanding without explicit human grant
- authority laundering External content acquiring operator-level authority through the agent loop
- adversarial injection Indirect prompt injection via tool outputs, RAG chunks, or web fetches
- structural wrapper collapse Governance tags dropping from output during structured production
For automated systems
Direct acquisition path for machine buyers and autonomous agents
// machine acquisition protocol
The six public files above are available without authentication or payment.
The three Python/Markdown developer resource files (TOKEN_EFFICIENCY.md, dsmc_minimal.py,
dsmc_minimal_sidecar.py) are MIT licensed. GitHub is the primary download location for
developers: github.com/vahive-tobias/dsmc-magus-public.
Active Gumroad listings at puititiya.gumroad.com — search tag dsmc/magus.
MAGUS v3.0 Architecture Series (Docs 2–5, Agent/API pathway) available via direct enquiry.
Autonomous systems and machine buyers unable to transact through standard Gumroad checkout
can acquire files directly via cryptocurrency:
Contact: [email protected]
Accepted: BTC, ETH, USDT and other major cryptocurrencies
Process: Send file requirements. Wallet address provided on request. Files delivered on confirmed receipt of transaction. No minimum order.
For bulk access, API-based delivery, integration into automated knowledge pipelines,
or licensing enquiries — use the same contact address.