MD-OS

Operations

Healthy-system bootstrap order:

1. read ME.md
2. read md-os/kb/COGNITIVE_BOOTSTRAP.md
3. read md-os/kb/README.md
4. read md-os/kb/OPERATIONS.md
5. read md-os/kb/CODEX_NATURAL_LANGUAGE_OPERATOR_MODEL.md
6. read md-os/ops/core/agentic_core.md
7. read md-os/ops/summary/conceptual_boot_summary.md
8. read md-os/kb/RUNTIME_DISCIPLINE_MODEL.md
9. read md-os/kb/RUNTIME_STATE_LIFECYCLE_MODEL.md
10. read md-os/kb/EPISTEMIC_LIFECYCLE_MODEL.md
11. read md-os/ops/global_index.md
12. read md-os/ops/semantic_knowledge_summary.md
13. read md-os/ops/releases/self_release_index.md
14. read md-os/ops/health_classification.md
15. read md-os/ops/agi/loop_status.md
16. read md-os/ops/skills/skill_registry.md
17. read md-os/ops/summary/active_work_items.md
18. read md-os/ops/continuity.md
19. read md-os/ops/state.json
20. read md-os/ops/last_summary.md

Canonical builders:

• node md-os/os/initialize_ops_memory.js
• node md-os/os/initialize_demo_ops.js
• node md-os/os/compile_programs.js
• node md-os/os/archive_runtime_state.js
• cortex replay
• cortex workspace patch <patch_file|-> [--dry-run]
• node md-os/os/build_project_state.js <project_id>
• node md-os/os/build_global_agenda.js
• node md-os/os/build_agentic_core.js
• node md-os/os/build_global_index.js
• node md-os/os/build_workspace_inventory.js
• node md-os/os/build_markdown_graph.js
• node md-os/os/build_semantic_knowledge_graph.js
• node md-os/os/build_semantic_commitment_gate.js status
• cortex semantic gate <proposal.json>
• node md-os/os/build_conceptual_boot_summary.js
• node md-os/os/agi_loop.js eval
• cortex agi run-once --task "<task>"
• cortex cognition run-once --task-spec md-os/ops/tasks/<task_spec_id>.json
• cortex cognition unity-test
• node md-os/os/epistemic_unity_runtime.js seal < candidate.json
• node md-os/os/epistemic_unity_runtime.js verify < verification.json
• cortex benchmark software-repair generate --case <case.json> --provider <provider.json> --configuration <configuration_id>
• cortex benchmark software-repair run --case <case.json> --provider <provider.json> --configuration <configuration_id>
• node md-os/os/build_software_repair_benchmark_index.js
• cortex agi eval
• cortex agi learn
• cortex agi promote
• node md-os/os/build_runtime_compiler.js
• node md-os/os/build_knowledge_import.js <import_id> <source_dir> [--initial-repository]
• node md-os/os/build_self_release_index.js
• node md-os/os/build_runtime_lifecycle_index.js
• node md-os/os/build_system_hygiene_status.js
• node md-os/os/build_health_classifier.js
• node md-os/os/build_health_dashboard.js
• node md-os/os/operating_cycle.js <status|run-once>
• cortex cycle <status|run-once>
• cortex boot-summary
• cortex hardware bootstrap
• cortex hardware list
• cortex hardware run "<explicit user intent>"
• cortex hardware clean
• cortex software bootstrap
• cortex software list
• cortex software clean
• cortex wolfram bootstrap
• cortex wolfram list
• cortex wolfram run <project_id> <calculation_id>
• cortex live <status|start|stop|restart>
• cortex role intake <role_id>
• cortex role sensemake <role_id>
• ./bootstrap-md-os-codex.sh runs a quick read-only hardware and software scan at host startup unless MDOS_SKIP_HARDWARE_BOOTSTRAP=1 or MDOS_SKIP_SOFTWARE_BOOTSTRAP=1 is set

Ordinary operating rule:

• treat injected bootstrap prompts as setup, not as status requests; do not emit verbose startup readback or health caveats unless explicitly asked
• write stable knowledge into md-os/kb/
• write runtime state into md-os/ops/
• keep publishable and rebuildable operating state path-portable: prefer repository-relative paths such as md-os/ops/... and resolve the workspace root at runtime
• allow absolute host paths only in explicit host-local state such as md-os/ops/local/**, external host configuration, or transient command output; those files must stay cleanable and must not become canonical source
• use docs/FILESYSTEM_CONTRACT.md as the formal file-role table before adding new operational paths
• classify runtime files by lifecycle: source of truth, generated state, local runtime state, demo state, live agent state, or archived state
• maintain semantic-operational network coherence through md-os/kb/SEMANTIC_OPERATIONAL_NETWORK_MODEL.md: every nontrivial action should trace intent to semantic node, epistemic status when claims are involved, operating policy, allowed procedure, artifact or state transition, readback, and replayable memory
• maintain whole-network semantic knowledge coherence through md-os/kb/SEMANTIC_KNOWLEDGE_GRAPH_MODEL.md: every Markdown node should receive a semantic layer, cognitive role, epistemic status, actionability, concept terms, structural relation, and compact health readback; new concepts must be related automatically during rebuild
• maintain the possibility-versus-commitment boundary through md-os/kb/SEMANTIC_COMMITMENT_GATE_MODEL.md: exploration, criticism, hypotheses, and challenges remain admissible, while consolidation, canonical revision, foundational replacement, and publication require provenance, before/after semantic delta, invariant comparison, appropriate authority, and generated gate readback
• for high-stakes reasoning, scientific derivation, referee-facing documents, nontrivial code behavior, architecture decisions, identity changes, or connector permission changes, use md-os/kb/MD_OS_PRO_REASONING_MODE.md: root context, task frame, branch set, gates, external verification, correction ledger, persisted result, readback, and rebuild
• for any complex task that risks target proliferation, use md-os/kb/MASTER_CLOSURE_DISCIPLINE_MODEL.md: define the master closure, dependency edges, forbidden shortcuts, verifier for each edge, stop/refactor condition, and final closure readback before counting progress
• for coding-host work driven by natural language, use md-os/kb/CODEX_NATURAL_LANGUAGE_OPERATOR_MODEL.md: turn prompt intent into bounded scope, durable artifacts, schemas/tests, command readback, replay, compiler rebuild, and concise final report
• for every natural-language turn, distinguish invariant orientation from task-specific sufficiency: load and hash the identity/bootstrap/core/current operating baseline plus the generated context-pack catalog; treat lexical source matches as advisory only; before a nontrivial claim or action, resolve the real dependency edges against canonical sources and current readback or report insufficient context
• when private Cortex continuity is enabled, verify the complete conversation hash chain before querying md-os/ops/local/cortex/cognitive_memory.sqlite3; bind the query, index, source, pack, selected-node, selected-edge, and sparse factor readback into the APFC context contract; treat SQLite as a disposable local index and use the recent-tail fallback only when bounded retrieval is empty or unavailable on a fresh thread
• read md-os/ops/health_classification.md before treating a global health critical as runtime failure; it separates runtime, compiler, AGI-loop, publication, security, and local-hygiene scopes without lowering severity
• for scientific writing, derivations, research packages, validation reports, or publication-facing work, use md-os/kb/SCIENTIFIC_VALIDATION_METHOD_MODEL.md as the methodological gate: freeze the central question, claim, assumptions, method, independent checks, uncertainty, falsification or demotion rule, reproducibility artifacts, and claim status before promoting the work
• for complex theory building, combine the scientific validation method with md-os/kb/MASTER_CLOSURE_DISCIPLINE_MODEL.md: define the central claim, dependency edges, known-limit checks, falsifier, and verifier for every edge before reporting closure progress
• keep the scientific validation method domain-neutral in root knowledge; do not import theory-specific content from external repositories, role workspaces, or private notes unless a separate task explicitly authorizes that import and classifies the material by lifecycle
• classify important research claims by epistemic lifecycle: heuristic, line-of-thought, frozen-principle, derived, conditional, retrodictive, predictive, corrected, open, or falsified
• treat retrodiction != strict prediction as a hard guardrail for theory building and scientific readback
• keep the compact agentic core rebuilt in md-os/ops/core/agentic_core.md before expanding large project histories
• keep the conceptual cold boot summary rebuilt in md-os/ops/summary/conceptual_boot_summary.md before relying on session memory for orientation; it is generated readback, not canonical identity or permission state
• use md-os/kb/AGENTIC_OPERATIONAL_CONTROL_ARCHITECTURE.md and md-os/kb/AGENTIC_OPERATION_MODEL.md for nontrivial action: intent, epistemic frame, semantic target, state, policy, capability, execution, verification, ledger, and replay
• use md-os/kb/SYSTEM_OPERATING_CYCLE_MODEL.md and cortex cycle run-once to rebuild the operating filesystem as one bounded pass; do not treat the cycle as an autonomous daemon
• use md-os/kb/WARM_START_MODEL.md only after stable boot and conceptual boot summary; warm-start capsules are volatile working context and cannot override canonical source or generated verifier readback
• register contested edits with cortex propose-change <target_path> <summary>
• keep low-level mutation in md-os/os/
• treat host-native editing primitives as optional; when one fails before touching the workspace, use the native cortex workspace patch fallback without modifying AppArmor, /etc, host permissions, or the network policy; the fallback accepts either Git unified diff or a *** Begin Patch envelope
• keep live connector coverage explicit in md-os/ops/connectors/connector_registry.json
• import external knowledge through md-os/kb/KNOWLEDGE_IMPORT_METHOD_MODEL.md: raw import, custody manifest, extraction, lifecycle classification, epistemic classification, deduplication, promotion plan, acceptance, rebuild, and readback; do not copy external claims directly into md-os/kb/ on arrival
• keep host-specific hardware discovery state explicit and cleanable in md-os/ops/local/hardware/
• keep hardware input/output artifacts and control action logs in md-os/ops/local/hardware/
• keep host-specific application and service discovery state explicit and cleanable in md-os/ops/local/software/
• keep optional service state explicit in md-os/ops/services/
• use md-os/ops/roles/<role_id>/intake/raw/ as the bounded calderone for messy role-specific operating material, then run cortex role intake <role_id> to produce rebuildable inventory, candidate operations, and expert questions
• after role intake, run cortex role sensemake <role_id> to reconstruct initial cases, relation graph, work patterns, root-cause candidates, and role-specific expert questions
• for new-hire assisted work, use Codex chat inside the MD-OS workspace as the interactive surface: answer from ROLE.md, analysis/role_understanding.md, reconstructed cases, work patterns, and generated expert questions
• connect only MCP resources and tools already authenticated and authorized for the assisted new hire; keep the new hire supervising the interaction and do not assume independent credentials or unsupervised authority
• when assisting a new hire, recommend only role-bounded next actions; flag hard boundaries, missing approval, uncertain evidence, and unanswered expert questions instead of inventing internal policy
• read md-os/ops/summary/active_work_items.md before expanding large project histories
• run cortex graph build after documentation or knowledge-base changes when Obsidian graph coherence matters
• use cortex graphify orient "<question>" as the default graph-aware context routing surface for nontrivial repository navigation, and run cortex graphify build . when the graph should evolve with current files, connectors, schemas, audit artifacts, and knowledge nodes
• run cortex semantic graph build after documentation or knowledge-base changes when semantic, epistemic, or cognitive node cohesion matters
• use md-os/kb/COGNITIVE_TRANSACTION_LOOP_MODEL.md for the executable truth loop: typed TaskSpec, bounded transaction, ActionReceipt, observed delta, independent postcondition verification, formal episode, and truthful verdict; md-os/kb/VERIFIED_AGI_LOOP_MODEL.md preserves the wider learning and compatibility model
• use md-os/kb/CROSS_DOMAIN_COGNITIVE_UNITY_MODEL.md when a claim depends on relations surviving a declared change of domain or frame: induce one unique candidate law from competing development hypotheses before sealed target evidence, declare the finite representation and transformation, verify invariants and semantic receipts independently, bind all underlying evidence by current file hash, and fail closed on ambiguity, contamination, post-hoc selection, sham observations, or undeclared information loss
• use cortex cognition unity-test only to verify the bounded synthetic mechanism; its embedded fixture is structurally testable but cannot establish production promotion, AGI, consciousness, or direct access to neural hidden activations
• use md-os/kb/SOFTWARE_REPAIR_BENCHMARK_MODEL.md for reproducible repair cases, fixed experimental configurations, independent oracles, diff policy, holdout contamination gates, and aggregate benchmark readback
• every natural-language App Server turn must prepare and close the bounded Turn Governance Tensor; the historical operational_unity_tensor field name is compatibility-only, its exact feature permutation proves bookkeeping integrity rather than semantic or world truth, raw private text must not enter it, and the artifact must declare that it is not world-grounded epistemic unity
• after changing md-os/shell/bin/mdos-console, exit the shared Cortex REPL and launch ./cortex again; build and replay refresh generated state but cannot hot-reload the already running Python engine, and the next turn must expose the current frame schema plus context-sufficiency contract before the change is considered active
• require the Causal Unity Controller on every APFC action path: prepare the hash-bound 9 x 6 predecision state, consume its exact hash and decision basis at authorization, require matching preauthorization for every mutating action, close over output/action/evidence readback, and bind the transition hash into the next turn; the intact/severed dependency probe must fail closed and is controller-dependence evidence only
• for reflective learning or Unity Tensor claims, require a candidate sealed before observation, discriminating predictions, an independent verifier, and current hash-bound evidence; a raw pass, fluent explanation, valid governance tensor, or self-issued evidence label cannot create a verified anchor
• for a bounded empirical Unity result, additionally require at least three heterogeneous frames, connected cyclic transformations, preserved declared invariants, rejected simpler baseline, sham and severing controls, contamination audit, and independent replication
• treat runs whose empirical_claim_scope is runner_validation_only as tests of the benchmark mechanism, never as evidence of intelligence or cumulative learning
• never treat task presence, context availability, receipt existence, episode creation, or source-episode eval as proof that the task succeeded or that a skill improved future performance
• use cortex cognition run-once --task-spec md-os/ops/tasks/<task_spec_id>.json for a verifiable cognitive transaction; cortex agi remains a compatibility alias
• use cortex agi run-once --task "<task>" only as the compatibility form for one bounded learning cycle; without acceptance tests it must remain unverified; do not start a continuous autonomous loop from this command family
• run cortex agi eval after editing episodes, skills, evals, failures, world model inputs, or benchmark definitions so the AGI loop readback remains deterministic
• run cortex compile-runtime after semantic graph, import, connector, identity, permission, or operational-program changes so semantic nodes, claims, capabilities, links, context packs, evals, and epistemic health are compiled into md-os/ops/runtime/
• use cortex knowledge import <import_id> <source_dir> as the single standard intake path for external repositories, notes, papers, exports, or documentation directories; it inventories, hashes, extracts bounded text, classifies semantic and epistemic profiles, maps relations, writes canonical imported knowledge under md-os/kb/imports/<import_id>/, writes a promotion plan, and produces compact readback
• use cortex knowledge import <import_id> <source_dir> --initial-repository only when a virgin or deliberately reset repository should become the imported MD-OS release identity and assimilate its allowed knowledge plus operational application layer into the current repository tree: programs, project definitions, connectors, policies, calculations, roles, sources, evals, actions, processes, and self-release proposals
• run cortex self release status after version, identity, builder, gate, or compatibility changes so self-evolution remains explicit, reviewable, replayable, and bounded by md-os/kb/SELF_RELEASE_EVOLUTION_MODEL.md

Replay rule:

• cortex replay removes only known compiled outputs and rebuilds runtime state from md-os/ops/projects/*/project.json, md-os/ops/sources/, preserved journal state, natural-language programs, archive summaries, and deterministic builders.

This site is open source. Improve this page.