Skip to content
View devdiv07's full-sized avatar

Highlights

  • Pro

Block or report devdiv07

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
devdiv07/README.md

Divyansh Shukla

Final-year CSE (AI) student working on reliability, open-source verification, and hardware-attestation testing.

I like engineering questions where a green test suite is not enough: what evidence supports the claim, what would falsify it, and does the implementation still hold when the input comes from somewhere independent?


AgenTrust work

Hardware-attestation verification

Recent work across the AgenTrust repositories has focused on verifier correctness and evidence provenance.

  • Reproduced a historical Intel TDX DCAP-v4 parser/fixture defect and isolated the difference to the six-byte QE_REPORT_CERTIFICATION_DATA wrapper.
  • Compiled Intel's upstream QVL quote generator and used independently originated evidence to reproduce the structural distinction without physical TDX hardware.
  • Generated TPM quote vectors with the Microsoft/TCG TPM 2.0 reference simulator and measured a signature/hash-agility false reject in a pinned Agent Manifest revision.

Contributions

cMCP #528
Delegate duplicate Intel TDX signature-section parsing to Agent Manifest's canonical parser, with regressions for overstated declared lengths.

cA2A #120
Remove redundant delegation-chain verification while preserving trusted-root, holder-proof, and authorization ordering.

Agent Manifest #323
Simulator-driven reproduction, verifier fix, and reference-simulator regression vectors for TPM signature/hash agility. Maintainer PR #320 subsequently landed the canonical implementation.


Selected engineering work

Durable execution for agent-initiated financial operations across retries, crashes, concurrency, and uncertain provider outcomes.

  • Separates logical operation identity from execution attempts and provider retry identity so retry safety is explicit rather than assumed.
  • Measured Razorpay Test Mode behavior and backed the failure model with crash, concurrency, authorization-replay, and mutation tests.
  • Public claims map to reproducible evidence rather than inferred API behavior.

A notify / digest / mute decision system with leakage-controlled evaluation, explicit policy precedence, and 277 automated tests.


Other open source

Razorpay MCP #114
Adds optional refund idempotency-key support to create_refund while preserving existing behavior when omitted.


Working with

Python · pytest · mutation testing · Git / GitHub · Docker · PostgreSQL · SQLAlchemy · REST APIs · TPM 2.0 / Intel TDX attestation work · LLM/VLM integration and evaluation


LinkedIn · Portfolio · Email

Final-year CSE (AI) · Jaipur

Pinned Loading

  1. agent-manifest agent-manifest Public

    Forked from agentrust-io/agent-manifest

    Agent Manifest SDK. Hardware-anchors all 10 artifacts defining an agent at deployment. Python and TypeScript.

    Python 1

  2. financial-operation-core financial-operation-core Public

    Durable execution for agent-initiated financial operations across retries, crashes, and uncertain provider outcomes.

    Python 1

  3. ca2a ca2a Public

    Forked from agentrust-io/ca2a

    Confidential agent-to-agent: attested, attenuated delegation and sealed peer channels as a profile on A2A

    Python 1

  4. fincore-chaos-lab fincore-chaos-lab Public

    Python 1

  5. cmcp cmcp Public

    Forked from agentrust-io/cmcp

    cMCP: Confidential MCP Gateway. Hardware-attested policy enforcement for MCP tool calls.

    Python 1

  6. ClaimTrace ClaimTrace Public

    Built a multimodal claim-verification pipeline where a vision LLM (gpt-4o) returns Pydantic-validated structured output (no freeform text), wrapped in deterministic rule checks and strict 14-column…

    Python 1