Skip to content

feat(memory-sentinel): add MCM governed agentic-memory slice - #267

Draft
Tarik Skalić (tarikskalic33) wants to merge 43 commits into
mainfrom
feat/cockroach-memory-sentinel-v1-final
Draft

feat(memory-sentinel): add MCM governed agentic-memory slice#267
Tarik Skalić (tarikskalic33) wants to merge 43 commits into
mainfrom
feat/cockroach-memory-sentinel-v1-final

Conversation

@tarikskalic33

@tarikskalic33 tarikskalic33 commented Aug 16, 2026

Copy link
Copy Markdown
Member

AEGIS Memory Sentinel — CockroachDB × AWS cash-conversion build

Status: DRAFT / DO NOT MERGE

Exact topology

  • base: main@32b7eb6a37fb69d19dd80189390b6641c5004ef1
  • head: 38a5cecda40182966dbdaebcf1904bc11f774aab
  • merge base: 32b7eb6a37fb69d19dd80189390b6641c5004ef1
  • ahead: 43
  • behind: 0
  • commits: 43
  • changed files: 32
  • additions: 1,858
  • deletions: 0
  • scope: all changed files are under hackathons/cockroach-memory-sentinel/
  • mergeable: TRUE at this snapshot

What is implemented

  • MCM — Mycorrhizal Collective Metacognition as a bio-inspired sparse meta-layer only; no claim that fungal/forest networks possess metacognition.
  • MCM hard boundary: OBSERVATION_ONLY / T2, authorityWeight=0, mayGroundStateTransition=false, no authority expansion.
  • deterministic memory-authority gate: stale state, stale policy, stale authority epoch, replay and missing binding all fail closed.
  • CockroachDB transactional state + VECTOR(1536) evidence-memory schema, vector index design, parameterized store adapter and real-engine receipt harness.
  • pinned official cockroachdb-sql Agent Skill provenance: cockroachlabs/cockroachdb-skills@e14e86d23ce8ee2e7e40a34ce2944c2502b6eadd, blob 2690e972a99fe632818f0fc1a434080bc7acd917.
  • OpenAI Agents SDK runtime (@openai/agents 0.16.0) with bounded tools: observe_collective_state, search_evidence, evaluate_action_memory.
  • agent constitution preserves DENY / NOT_ESTABLISHED; model/retrieval/MCM output is evidence, never authority.
  • AWS Lambda + SAM nodejs24.x deployment path, bounded concurrency, public no-spend /health, bearer gate before POST model/DB allocation.
  • live deployment verifier refuses to pass unless an actual Agents SDK run records evaluate_action_memory.
  • secret-safe run evidence stores tool names/count only, not raw tool arguments or bearer tokens.
  • full Devpost field map and execution-to-submission runbook.

Combined local component witness — promoted

evidence/combined-local-component-witness.json

A single TAP execution was run after verifying exact file identity with git hash-object against the GitHub blob SHAs for 8 source + 8 test files.

Result:

  • tests: 30
  • pass: 30
  • fail: 0
  • skipped: 0
  • cancelled: 0
  • TAP output SHA-256: 46d59cc8feeee097da72fa55fc4d8d8c6ca8be70616246e431a95b368c390d60
  • witness SHA-256: 084b6997ac5365a428b4b59e99be05c7fd03b7b8f0151deef3b130a59983435c

Execution environment was Node v22.16.0; declared Lambda runtime is Node 24, therefore RUNTIME_PARITY_WITH_DECLARED_NODE24 = NOT_ESTABLISHED.

This promotes only:

  • EXACT_TESTED_BLOB_BINDING: ESTABLISHED_FOR_LISTED_16_BLOBS
  • COMBINED_LOCAL_COMPONENT_TEST_PASS: ESTABLISHED_30_OF_30

It does not promote full-subtree exact-head testing, dependency-bearing OpenAI execution, Cockroach runtime, AWS runtime or production admission.

Current evidence ledger

  • MCM_CONTRACT_IMPLEMENTATION: ESTABLISHED_IN_BRANCH
  • MEMORY_AUTHORITY_GATE: ESTABLISHED_IN_BRANCH
  • COCKROACH_VECTOR_STORE_ADAPTER: ESTABLISHED_IN_BRANCH
  • COCKROACH_SCHEMA: IMPLEMENTED
  • COCKROACH_AGENT_SKILL_PROVENANCE_BINDING: ESTABLISHED
  • OPENAI_AGENT_RUNTIME_CODE: IMPLEMENTED
  • OPENAI_TOOL_CALL_EVIDENCE_SURFACE: IMPLEMENTED
  • AWS_LAMBDA_DEPLOYMENT_CODE: IMPLEMENTED
  • AWS_RUNTIME_RECEIPT_HARNESS: IMPLEMENTED
  • COMBINED_LOCAL_COMPONENT_TEST_PASS: ESTABLISHED_30_OF_30
  • NODE24_RUNTIME_PARITY: NOT_ESTABLISHED
  • DEPENDENCY_BEARING_OPENAI_RUNTIME_PASS: NOT_ESTABLISHED
  • COCKROACHDB_REAL_ENGINE_RUNTIME_RECEIPT: NOT_ESTABLISHED
  • COCKROACHDB_DISTRIBUTED_VECTOR_INDEX_EXECUTION: NOT_ESTABLISHED
  • COCKROACHDB_AGENT_SKILL_EXECUTION: NOT_ESTABLISHED
  • OPENAI_LIVE_AGENT_INFERENCE: NOT_ESTABLISHED
  • AWS_LAMBDA_LIVE_DEPLOYMENT: NOT_ESTABLISHED
  • FUNCTIONAL_PUBLIC_DEMO: NOT_ESTABLISHED
  • HACKATHON_STAGE_ONE_ELIGIBILITY: NOT_YET_ESTABLISHED

Pre-existing-work disclosure

AEGIS Omega predates this challenge and already contains governance, replay, receipt and local metacognitive experiments. The challenge-period work is the new hackathons/cockroach-memory-sentinel/ subtree: MCM collective layer, persistent-memory authority contract, CockroachDB vector/store path, sponsor-skill binding/verifiers, bounded OpenAI runtime, AWS deployment/verification path, tests, runbook and submission artifacts.

Remaining admission gates before Devpost submit

  1. dependency-bearing npm install under Node 24 and preserve package-lock.json;
  2. run npm run verify:cockroach against a real CockroachDB 25.4+ cluster;
  3. execute the pinned Cockroach Agent Skill against the connected cluster and preserve SQL/EXPLAIN evidence;
  4. deploy SAM app to AWS Lambda;
  5. complete secure OpenAI project-key setup and run live agent inference;
  6. run npm run verify:aws and require live evaluate_action_memory evidence;
  7. record the required public <3 minute video showing the Cockroach memory layer;
  8. provide functional demo URL and personally confirm the Devpost legal checkboxes;
  9. bind final submission to the exact final Git SHA.

No sponsor integration, live cloud execution, production, certification, AGI, novelty or revenue claim is made before those receipts exist.

@vercel

vercel Bot commented Aug 16, 2026

Copy link
Copy Markdown

Deployment failed for project hub with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/tariks-projects-e9198507?upgradeToPro=build-rate-limit

@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Aug 16, 2026

Copy link
Copy Markdown
Contributor

Deploying with  Cloudflare Workers  Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status Name Latest Commit Preview URL Updated (UTC)
✅ Deployment successful!
View logs
aegisomega 38a5cec Commit Preview URL

Branch Preview URL
Aug 16 2026, 08:15 AM

@vercel

vercel Bot commented Aug 16, 2026

Copy link
Copy Markdown

Deployment failed for project hook-generator with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/tariks-projects-e9198507?upgradeToPro=build-rate-limit

@vercel

vercel Bot commented Aug 16, 2026

Copy link
Copy Markdown

Deployment failed for project platform-picker with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/tariks-projects-e9198507?upgradeToPro=build-rate-limit

@vercel

vercel Bot commented Aug 16, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
hook-generator Ready Ready Preview Aug 16, 2026 8:10am

@vercel

vercel Bot commented Aug 16, 2026

Copy link
Copy Markdown

Deployment failed for project hook-generator with the following error:

Resource is limited - try again in 1 day (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/tariks-projects-e9198507?upgradeToPro=build-rate-limit

@vercel

vercel Bot commented Aug 16, 2026

Copy link
Copy Markdown

Deployment failed for project hub with the following error:

Resource is limited - try again in 1 day (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/tariks-projects-e9198507?upgradeToPro=build-rate-limit

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant