Spaces:
Running
Running
File size: 19,631 Bytes
518343a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 | // doctrine-scanner-exempt: legacy live-product surface; rename tracked as separate engineering debt — see scripts/check-doctrine-v6.mjs header.
// ARGO — Field Intelligence Forge
// Integration of external open-source primitives + frontier public research
// + the Sentra defensive doctrine into capability seeds for the
// A11oy / Sentra / Psyche / Argo constellation.
//
// External code is studied with-knowledge-of, not adopted. Frontier methods are
// reimplemented from first principles using only public inputs. Adoption requires
// a Sentra approval workflow per the Glasswing doctrine.
export const ARGO_VERSION = '0.2.0-seed';
export const ARGO_TAGLINE =
'External signal → integrated primitive → constellation evolution. Defensive only, evidence-bound, human-gated, public-input only.';
// ---------------------------------------------------------------------------
// 1. DOCTRINE PILLARS
// Quoted from the Sentra Defensive Executive Summary and the Mythos/Glasswing
// research brief. Every seed must satisfy at least one pillar and violate none.
// ---------------------------------------------------------------------------
export type DoctrinePillarId =
| 'no-hack-back'
| 'human-in-loop'
| 'evidence-is-evidence'
| 'detection-targets'
| 'public-inputs-only';
export interface DoctrinePillar {
id: DoctrinePillarId;
pillar: string;
citation: string;
quote: string;
enforcedBy: string;
}
export const DOCTRINE_PILLARS: readonly DoctrinePillar[] = [
{
id: 'no-hack-back',
pillar: 'No hack-back. Ever.',
citation: '18 U.S.C. §1030 (CFAA)',
quote:
'Any unauthorized access or damaging code is a felony. Even Texas property-defense law does not legalize hacking outside one\u2019s network.',
enforcedBy: 'Cerberus hard boundary · objective-function constraint',
},
{
id: 'human-in-loop',
pillar: 'AI is a recommender, not an autonomous actor.',
citation: 'NIST AI Risk Management Framework',
quote:
'For any non-trivial action above a low-risk threshold, the AI\u2019s recommendation triggers a human approval workflow.',
enforcedBy: 'Sentra Approval Queue · dual-key gating',
},
{
id: 'evidence-is-evidence',
pillar: 'Collected incident data is evidence.',
citation: 'NIST SP 800-61r3',
quote:
'Even if chain-of-custody is not invoked for every case, all incident data must be preserved with provenance suitable for regulators or law enforcement.',
enforcedBy: 'Cerberus Evidence Vault · append-only, hash-linked',
},
{
id: 'detection-targets',
pillar: 'MTTD <1h. MTTR <4h.',
citation: 'SOC industry benchmark (executive summary §8)',
quote:
'Reduce Mean Time To Detect to under 1 hour and Mean Time To Respond to under 4 hours for major incidents. Track MITRE ATT&CK coverage.',
enforcedBy: 'Silver RL planner · reward shaping favors fast verified containment',
},
{
id: 'public-inputs-only',
pillar: 'Public inputs only. No leaks, no scraped weights.',
citation: 'Mythos doctrine — Ethical/Legal/IP Considerations',
quote:
'We rely solely on published research, official repos, and licensed data. Reverse-engineering proprietary weights or using leaked artifacts is forbidden.',
enforcedBy: 'Hephaestus provenance gate · SLSA-style supply-chain attestation',
},
];
// ---------------------------------------------------------------------------
// 2. CAPABILITY SEEDS — integrated from external sources
// ---------------------------------------------------------------------------
export type SeedStatus = 'active' | 'integrated' | 'adoptable' | 'piloted';
export type ConstellationTarget = 'A11oy' | 'Sentra' | 'Psyche' | 'Argo';
export type RoadmapPhase = '0\u20136' | '7\u201312' | '13\u201324';
export type SeedSourceKind = 'open-source' | 'public-research';
export type SeedId =
// ei-grad open-source primitives
| 'kill-switch'
| 'edge-distill'
| 'oidc-attest'
| 'cert-hygiene'
| 'evidence-snapshot'
| 'kernel-tap'
| 'device-trust'
| 'observatory-repl'
// Anthropic public-research frontier methods
| 'skills-standard'
| 'system-card-practice'
| 'cot-monitorability'
| 'rdt-research'
| 'frontier-red-team';
export interface CapabilitySeed {
id: SeedId;
name: string;
oneLine: string;
primitive: string;
sourceKind: SeedSourceKind;
source: { repo: string; url: string; author: string; lang: string };
evolves: ConstellationTarget;
evolvesAgent: string;
doctrineLink: DoctrinePillarId;
defensiveGuardrail: string;
phase: RoadmapPhase;
status: SeedStatus;
}
export const CAPABILITY_SEEDS: readonly CapabilitySeed[] = [
// -------------------- OPEN-SOURCE PRIMITIVES (ei-grad) --------------------
{
id: 'kill-switch',
name: 'Deterministic agent kill-switch',
oneLine:
'Every Silver action wrapped in a hard wall-clock budget. SIGTERM at soft limit, SIGKILL at hard limit, with traceback preserved.',
primitive:
'multiprocessing.Process + tblib + SIGKILL escalator: a function decorator that guarantees termination regardless of internal state.',
sourceKind: 'open-source',
source: { repo: 'kill-timeout', url: 'https://github.com/ei-grad/kill-timeout', author: 'ei-grad/Andrew Grigorev', lang: 'Python' },
evolves: 'A11oy',
evolvesAgent: 'Pallas (planner-hardener) + Silver (RL executor)',
doctrineLink: 'human-in-loop',
defensiveGuardrail:
'Operates only on the agent\u2019s own process tree; no remote signals, no escalation outside the worker.',
phase: '0\u20136',
status: 'adoptable',
},
{
id: 'edge-distill',
name: 'Edge defender via knowledge distillation',
oneLine:
'Distill the heavy detection model into a tiny on-host classifier that runs on offline endpoints. Sentra without the cloud round-trip.',
primitive:
'Teacher\u2192student model compression for inference at the edge of a sparse network.',
sourceKind: 'open-source',
source: { repo: 'samogonka', url: 'https://github.com/ei-grad/samogonka', author: 'ei-grad/Andrew Grigorev', lang: 'Python' },
evolves: 'Sentra',
evolvesAgent: 'Sentinel (endpoint defender)',
doctrineLink: 'detection-targets',
defensiveGuardrail:
'Edge student is read-only at the edge; can only emit alerts upstream, never execute remediation locally without Sentra approval.',
phase: '7\u201312',
status: 'integrated',
},
{
id: 'oidc-attest',
name: 'OIDC-attested action provenance',
oneLine:
'Every defensive action signed by an OIDC-verified agent identity. Token + audience + issuer captured as an inline attestation in the evidence record.',
primitive:
'Strict OIDC token verification (issuer, audience, kid, exp) at the action boundary, before the policy engine fires.',
sourceKind: 'open-source',
source: { repo: 'verify-oidc-token', url: 'https://github.com/ei-grad/verify-oidc-token', author: 'ei-grad/Andrew Grigorev', lang: 'Python' },
evolves: 'Sentra',
evolvesAgent: 'Hephaestus (forge / signing) + Cerberus (vault)',
doctrineLink: 'evidence-is-evidence',
defensiveGuardrail:
'Attestation is a passive witness, not an authorisation. Approval queue still gates the action.',
phase: '13\u201324',
status: 'adoptable',
},
{
id: 'cert-hygiene',
name: 'Pre-emptive cert expiry scanner',
oneLine:
'Sentra continuously checks customer asset registry for HTTPS certs nearing expiry. A defence that fires before the incident.',
primitive:
'Fast Go scanner that reads cert expiration metadata from a host list and emits warnings on a configurable threshold.',
sourceKind: 'open-source',
source: { repo: 'check-expiring-certs', url: 'https://github.com/ei-grad/check-expiring-certs', author: 'ei-grad/Andrew Grigorev', lang: 'Go' },
evolves: 'Sentra',
evolvesAgent: 'Argus (perimeter watcher)',
doctrineLink: 'detection-targets',
defensiveGuardrail:
'Read-only TLS handshake against assets in the tenant registry. Never probes hosts outside the registered scope.',
phase: '0\u20136',
status: 'adoptable',
},
{
id: 'evidence-snapshot',
name: 'Hardlink-snapshot evidence vault',
oneLine:
'Minute-by-minute deduplicated snapshots of evidence using hardlinks. Storage stays bounded, chain-of-custody stays clean.',
primitive:
'rsync + hardlink snapshot rotation: unchanged files share inodes across snapshots, reducing storage to delta-only cost.',
sourceKind: 'open-source',
source: { repo: 'trinkup', url: 'https://github.com/ei-grad/trinkup', author: 'ei-grad/Andrew Grigorev', lang: 'Bash' },
evolves: 'Sentra',
evolvesAgent: 'Cerberus (evidence vault)',
doctrineLink: 'evidence-is-evidence',
defensiveGuardrail:
'Vault is append-only at the logical layer; snapshots augment, never overwrite, the immutable hash chain.',
phase: '13\u201324',
status: 'integrated',
},
{
id: 'kernel-tap',
name: 'Kernel-level connection telemetry',
oneLine:
'Per-flow netfilter conntrack as a raw signal source. Sees east-west traffic that EDR alone misses.',
primitive:
'libnetfilter_conntrack ctypes binding: emit connection state transitions to the SIEM ingest pipeline.',
sourceKind: 'open-source',
source: { repo: 'python-conntrack', url: 'https://github.com/ei-grad/python-conntrack', author: 'ei-grad/Andrew Grigorev', lang: 'Python' },
evolves: 'Sentra',
evolvesAgent: 'Argus + Sentinel (telemetry fusion)',
doctrineLink: 'detection-targets',
defensiveGuardrail:
'Read-only on the host\u2019s own connection table. No packet injection, no manipulation of conntrack state.',
phase: '7\u201312',
status: 'active',
},
{
id: 'device-trust',
name: 'EAP-TLS device trust membrane',
oneLine:
'Operator workstations and field sensors authenticate to the Sentra control plane with mutual TLS over RADIUS. No cert, no console.',
primitive:
'Minimal EAP-TLS FreeRADIUS image with single-CA + OCSP revocation. Strong device identity, no shared secrets.',
sourceKind: 'open-source',
source: { repo: 'docker-freeradius-eap-tls', url: 'https://github.com/ei-grad/docker-freeradius-eap-tls', author: 'ei-grad/Andrew Grigorev', lang: 'Dockerfile' },
evolves: 'Sentra',
evolvesAgent: 'Argus (zero-trust admission)',
doctrineLink: 'human-in-loop',
defensiveGuardrail:
'Authenticates only devices the founder has explicitly enrolled. Revocation propagates within OCSP TTL.',
phase: '7\u201312',
status: 'integrated',
},
{
id: 'observatory-repl',
name: 'Governed operator REPL into Psyche',
oneLine:
'IPython-style live console attached to a running Psyche observatory process. Read-only by default; mutation requires Sentra approval.',
primitive:
'Replace the framework\u2019s default shell with a richer interactive REPL bound to live application context.',
sourceKind: 'open-source',
source: { repo: 'flask-shell-ipython', url: 'https://github.com/ei-grad/flask-shell-ipython', author: 'ei-grad/Andrew Grigorev', lang: 'Python' },
evolves: 'Psyche',
evolvesAgent: 'Oracle (observability) + Hermes (operator channel)',
doctrineLink: 'human-in-loop',
defensiveGuardrail:
'REPL session is OIDC-attested, time-boxed by kill-switch budget, every command logged to the audit chain.',
phase: '13\u201324',
status: 'active',
},
// ---------- FRONTIER METHODS (Anthropic / OpenMythos public research) ----------
{
id: 'skills-standard',
name: 'Declarative skills with capability scopes',
oneLine:
'Plugin-style skills declared in a manifest with explicit capability scopes. Loaded into agent runtime under Sentra policy, never auto-elevated.',
primitive:
'Skill manifest pattern (anthropics/skills): name, description, allowed tools, signing authority. Agent picks skills only when policy permits.',
sourceKind: 'public-research',
source: { repo: 'anthropics/skills', url: 'https://github.com/anthropics/skills', author: 'Anthropic (public)', lang: 'Markdown / Python' },
evolves: 'A11oy',
evolvesAgent: 'Hermes (operator channel) + Pallas (plugin governance)',
doctrineLink: 'human-in-loop',
defensiveGuardrail:
'Every capability scope is a Sentra-policy permission. Skill load is auditable; capability elevation requires re-attestation in the approval queue.',
phase: '0\u20136',
status: 'integrated',
},
{
id: 'system-card-practice',
name: 'Public system card per release',
oneLine:
'No agent enters production rotation without a public capabilities + alignment risk report. Card is reviewed by Cerberus before promotion.',
primitive:
'System Card discipline: capabilities, evaluation results, known limitations, intended use, alignment risk addendum, all published with the release.',
sourceKind: 'public-research',
source: { repo: 'anthropic.com/research', url: 'https://www.anthropic.com/research', author: 'Anthropic (public)', lang: 'Markdown / PDF' },
evolves: 'A11oy',
evolvesAgent: 'Oracle (observability) + Daedalus (architect)',
doctrineLink: 'evidence-is-evidence',
defensiveGuardrail:
'Card review is a hard gate, not an after-the-fact disclosure. Missing or stale card blocks the agent from the runtime registry.',
phase: '7\u201312',
status: 'piloted',
},
{
id: 'cot-monitorability',
name: 'Chain-of-thought as first-class telemetry',
oneLine:
'Surface model reasoning steps to human auditors as a read-only telemetry stream. Treat the chain-of-thought as evidence, not as authorisation.',
primitive:
'Reasoning trace logging + diff against prior runs; flag drift, hidden-goal patterns, or evaluator-aware behaviour. Independently reimplemented from public methodology.',
sourceKind: 'public-research',
source: { repo: 'CoT monitorability research (public)', url: 'https://arxiv.org/abs/2407.13692', author: 'Mikulik, Bowman et al. (public preprint)', lang: 'Methodology' },
evolves: 'Psyche',
evolvesAgent: 'Oracle (observability) + Ariadne (memory)',
doctrineLink: 'evidence-is-evidence',
defensiveGuardrail:
'Chain-of-thought is read-only telemetry. Never used as an authorisation channel; never replayed back into the agent as an instruction.',
phase: '13\u201324',
status: 'integrated',
},
{
id: 'rdt-research',
name: 'Recurrent-depth + sparse MoE reference architecture',
oneLine:
'Iterative-compute transformer for hard reasoning: looped depth, switchable attention, sparse expert routing. Studied in clean-room only.',
primitive:
'Recurrent-Depth Transformer + MLA/GQA switchable attention + sparse Mixture-of-Experts feed-forward, per OpenMythos public reconstruction.',
sourceKind: 'public-research',
source: { repo: 'OpenMythos reference', url: 'https://github.com/anthropics', author: 'OpenMythos (independent OSS)', lang: 'Python' },
evolves: 'A11oy',
evolvesAgent: 'Daedalus (architect) + Twin Foundry',
doctrineLink: 'public-inputs-only',
defensiveGuardrail:
'Clean-room reimplementation only. No proprietary weights, no scraped private data. Trained only on permissively-licensed public corpora.',
phase: '13\u201324',
status: 'active',
},
{
id: 'frontier-red-team',
name: 'Internal frontier red-team gate',
oneLine:
'Dedicated internal red-team probes new agent capabilities before release. Findings flow into the Glasswing patch queue as P-tagged risk items.',
primitive:
'Standing red-team practice with autonomous-code-reasoning evaluations on a closed test fleet. Output: an alignment risk report attached to the system card.',
sourceKind: 'public-research',
source: { repo: 'Anthropic Frontier Red Team blog', url: 'https://www.anthropic.com/research', author: 'Anthropic (public)', lang: 'Methodology' },
evolves: 'A11oy',
evolvesAgent: 'Pallas (planner-hardener) + Sentinel (endpoint defender)',
doctrineLink: 'detection-targets',
defensiveGuardrail:
'Red-team operates only against Sentra-owned test fleets. Never against customer assets, never against third-party infrastructure.',
phase: '0\u20136',
status: 'piloted',
},
];
// ---------------------------------------------------------------------------
// 3. ALIGNMENT MATRIX — capability → satisfying seeds, with seed-id literal union
// ---------------------------------------------------------------------------
export interface AlignmentRow {
capability: string;
satisfiedBy: readonly SeedId[];
}
export const ALIGNMENT_MATRIX: readonly AlignmentRow[] = [
{ capability: 'EDR/EPP — endpoint detection', satisfiedBy: ['edge-distill', 'kernel-tap'] },
{ capability: 'SIEM/XDR — telemetry fusion', satisfiedBy: ['kernel-tap', 'cert-hygiene'] },
{ capability: 'SOAR — orchestrated response', satisfiedBy: ['kill-switch', 'oidc-attest'] },
{ capability: 'IAM/IdP — zero-trust identity', satisfiedBy: ['oidc-attest', 'device-trust'] },
{ capability: 'Evidence handling — chain of custody', satisfiedBy: ['oidc-attest', 'evidence-snapshot'] },
{ capability: 'Policy engine — human-in-the-loop', satisfiedBy: ['kill-switch', 'observatory-repl', 'skills-standard'] },
{ capability: 'Network controls — perimeter', satisfiedBy: ['cert-hygiene', 'device-trust'] },
{ capability: 'Skills / Plugins — extensibility under policy', satisfiedBy: ['skills-standard'] },
{ capability: 'Pre-release safety review', satisfiedBy: ['system-card-practice', 'frontier-red-team'] },
{ capability: 'Reasoning telemetry & monitorability', satisfiedBy: ['cot-monitorability'] },
{ capability: 'Reference architecture R&D', satisfiedBy: ['rdt-research'] },
];
// ---------------------------------------------------------------------------
// 4. CITATIONS
// ---------------------------------------------------------------------------
export const ARGO_CITATIONS: ReadonlyArray<{ label: string; url: string; kind: 'repo' | 'doctrine' | 'standard' }> = [
{ label: 'ei-grad — repository index', url: 'https://github.com/ei-grad?tab=repositories', kind: 'repo' },
{ label: 'anthropics/skills — Agent Skills standard (public)', url: 'https://github.com/anthropics/skills', kind: 'repo' },
{ label: 'Anthropic Research — system cards & frontier red team (public)', url: 'https://www.anthropic.com/research', kind: 'doctrine' },
{ label: 'OpenMythos — independent public reconstruction (RDT/MoE)', url: 'https://github.com/anthropics', kind: 'repo' },
{ label: 'Chain-of-Thought monitorability — Mikulik, Bowman et al. (public preprint)', url: 'https://arxiv.org/abs/2407.13692', kind: 'doctrine' },
{ label: 'NIST SP 800-61r3 — Computer Security Incident Handling Guide', url: 'https://csrc.nist.gov/pubs/sp/800/61/r3/final', kind: 'standard' },
{ label: 'NIST AI Risk Management Framework (AI RMF 1.0)', url: 'https://www.nist.gov/itl/ai-risk-management-framework', kind: 'standard' },
{ label: '18 U.S.C. §1030 — Computer Fraud and Abuse Act', url: 'https://www.law.cornell.edu/uscode/text/18/1030', kind: 'standard' },
{ label: 'DARPA AI Cyber Challenge (AIxCC)', url: 'https://aicyberchallenge.com/', kind: 'doctrine' },
];
|