Cybersecurity Engineer · AI Agent · Android/Termux Developer
Bengaluru-based engineer working across VAPT, SOC/SIEM, security automation, and AI systems that run on Linux, Android, and Termux — from enterprise security operations to open-source AI security tooling.
Cybersecurity engineer and software/AI engineer focused on VAPT, security automation, Linux/Termux engineering, Android automation, and AI agents.
Yashas Prakash is a Bengaluru-based cybersecurity engineer and developer. His work spans enterprise security operations, application security testing, security tooling, Linux systems engineering, and — more recently — AI agents and autonomous systems engineered onto Android and Termux. He has built open-source security and automation tools used across VAPT workflows, Android device automation, and agent-based browser control.
GitHub: yashas-13 · LinkedIn: Yashas Prakash · Company: Pravidhi Tech Solutions LLP
Enterprise security through engineering and leadership, then into productizing security and AI.
Pravidhi Tech Solutions LLP (LLPIN ACO-8103) · Bengaluru
Leads cybersecurity consulting, VAPT, security automation, AI engineering, software engineering, Android/Termux development, browser automation, and AI-agent systems. Builds open-source security tooling including Pentest-Pi, Marina, AccessDroid-Termux, Termux Coding Agent Toolkit, and BrowserClaw.
Enterprise cybersecurity role spanning:
Aug 2020 – Aug 2021 · 1 yr 1 mo
Grouped by domain so a client, recruiter, or journalist can immediately see where the work is credible.
| Domain | Technologies |
|---|---|
| Cybersecurity | VAPT · Web Security · API Security · Network Security · Cloud Security · SOC · SIEM · WAF · XDR/SOAR · PTES · OWASP |
| Systems | Linux · Kali Linux · Bash · Docker · Kubernetes · WSL · Termux · Ubuntu · Elasticsearch · Kibana · Fleet |
| Development | Python · Node.js · Git · REST APIs · Android · AccessibilityService · Incremental builds |
| AI | LLMs · AI Agents · VLMs · MCP · Claude · Codex · OpenCode · 9Router · Browser automation |
| Cloud | AWS · Azure · GCP · IAM · Terraform · Hardening · CIS Benchmark |
Linux is not just an OS here — it is the systems layer underneath SOC/SIEM and the base that Termux and later AI agents compose on.
Flagship builds first, then supporting engineering. Each line links to verifiable evidence (repository, post, or demo).
| Project | Domain | Evidence |
|---|---|---|
| Pentest-Pi | AI + Cybersecurity | GitHub |
| Marina | Android AI Agent | GitHub |
| AccessDroid-Termux | Android Automation | GitHub |
| Termux Coding Agent Toolkit | Developer Tooling | GitHub |
| BrowserClaw | Agent Automation | Company post |
Agent-driven security workflow that composes reconnaissance, scanning, vulnerability analysis, exploitation validation, and automated reporting with OWASP Top 10 mapping, security-tool orchestration, and Termux-native deployment. Built for authorized testing, labs, and CTFs.
Android agent architecture built on AccessibilityService and MediaProjection: screen understanding, visual interaction, Shizuku elevated-input fallback, voice, overlays, persistence, code editing, and project management. Fills the missing middle between AccessDroid (automation substrate) and BrowserClaw/Pentest-Pi (full autonomous workflows).
Android automation toolkit that lets a Termux agent control the Android UI through AccessibilityService: tap, scroll, long-press, swipe, type, app launch, UI-node selection, screenshots, UI-tree inspection. Adds a VLM loop (screenshot → VLM → action JSON → coordinate mapping → AccessibilityService → gesture) with configurable OpenAI-compatible endpoints including local models. Systems-level Android engineering, not scripting.
A Termux development environment documented with 150+ tools across 20+ categories covering programming languages, networking, databases, debugging, Android APIs, and AI tooling. Built as a Termux bootstrap that turns an Android device into a developer and security workstation.
Agent automation that receives a command in Telegram, drives a real browser via BrowserOS MCP, and posts to GitHub or LinkedIn through a company-owned identity gate. One stdlib-only Python file orchestrates MCP, Claude Code, OpenCode, and 9Router. Architecture: Telegram → Agent → LLM/Coding Agent → Browser → MCP → Website/GitHub/LinkedIn.
ISO 27001 ISMS and the policies and processes it entails: access control, incident management, supplier security, business continuity/DR, application and network security, and security-governance documentation. The work at this layer turns technical findings into auditable controls.
GitHub experiments and product work including Pravidhi-Flowy (AppFlowy-based), Pushpa Family Chit Fund, ParkZone, and the supporting substrate for AI agents (OpenClaw/ManyClaw, skill runtime SKILL.md → Agent → Read/Edit/Bash, 9Router/Claude/Codex orchestration, local-LLM/VLM exploration). Each is presented as engineering direction with public evidence where available, rather than exclusively as commercial launches.
Public repositories live at github.com/yashas-13 with active development across the flagship projects listed above. See each project’s repository for architecture docs, setup, and the latest releases. The portfolio’s closed-loop story is Security Engineering → Automation → AI Agents → Mobile/Android Computer Use, where Pentest-Pi sits at the convergence of all tracks.
Recent upstream engineering contributions to OpenClaw, including fixes that were reviewed, merged, superseded by equivalent fixes, or explicitly carried forward by maintainers.
| PR | Contribution | Maintainer acknowledgement |
|---|---|---|
| #155994 | Preserve stored Copilot Responses reasoning ciphertext with long provider IDs; added regression and live wire-to-SQLite integrity proof. | obviyus: “Thanks @yashas-13 for the contribution and the clear report.” Landed in 92b27867ca. |
| #157348 | Materialize the exact-session-entry environment before worker transport to prevent Windows DataCloneError. | steipete: “Thanks, @yashas-13, for the exact-entry environment Proxy investigation and proposed repair.” Finding carried forward to #160075. |
| #157366 | Preserve already-dispatched sibling tools across sessions_yield handoff cancellation boundaries. | Upstream agent/session lifecycle contribution. |
| #161103 | Preserve configured Codex max thinking capability on reply-path turns with production Gateway-to-harness proof. | obviyus: “Landed as 2fe1706154. Thanks @yashas-13!” |
| #161179 | Clarify episodic transcript recall via Active Memory and correct documentation formatting. | obviyus: “Thanks @yashas-13 for the documentation fix!” Merged. |
| #162390 | Share plugin metadata scope across multi-agent Doctor preparation to reduce retained config/registry duplication. | steipete: “Thanks @yashas-13” and confirmed the identified per-agent metadata-scope problem was covered by the fix landed on main. |
| #162401 | Preserve fenced blocks in newline chunk mode so channel adapters retain fence-aware chunking. | obviyus: “Thanks @yashas-13.” Closed as a duplicate after the upstream fix merged. |
| #162449 | Decode percent-encoded local JSON Schema references before JSON Pointer unescaping. | obviyus: “Thanks @yashas-13 for working on this.” Closed after the earlier equivalent fix merged. |
| #162457 | Close pnpm package-install stdin so updater subprocesses receive EOF instead of inheriting the caller TTY. | steipete: “Thanks @yashas-13” while explaining the equivalent fix already landed on main. |
| #163779 | Preserve Responses reasoning ciphertext when long reasoning identifiers pass through transcript/storage handling. | OpenClaw Responses/reasoning integrity contribution. |
| #162461 | Resolve percent-encoded local JSON Schema $ref fragments safely, including malformed encoding behavior. | Schema-resolution contribution; equivalent upstream fix tracked separately. |
| #162388 | Defer CLI-only bootstrap imports until after the --version fast path. | Startup/import-boundary performance and correctness contribution. |
| #162379 | Release captured async-work trackers when their captured scope begins closing, falling back to caller-owned execution. | Async-work lifecycle correctness contribution. |
| #162377 | Bound repeated Discord READY recovery attempts and escalate persistent startup failure to the provider supervisor. | Discord lifecycle/recovery reliability contribution. |
| #161146 | Reuse the configured workspace for system-agent planner turns to avoid unnecessary prepared-runtime cache churn. | Planner/runtime resource-efficiency contribution. |
| #161107 | Keep the audit writer alive after one malformed, non-cloneable event and continue accepting later valid events. | Audit durability/recovery contribution. |
| #161106 | Report foreign Gateway port holders promptly instead of waiting through the full restart-health budget. | Gateway diagnostics and operational reliability contribution. |
| #156273 | Preserve OKF concept type as top-level metadata during memory-wiki import with regression coverage. | Memory/import data-integrity contribution. |
| #135196 | Improve strict-JSON shell-quoting diagnostics, including PowerShell-safe structured-value guidance. | CLI diagnostics/documentation contribution. |
Direct maintainer comments acknowledge the work above. Peter Steinberger (@steipete) personally thanked Yashas on #157348, #162390, and #162457. Maintainer @obviyus separately thanked Yashas on #155994, #161103, #161179, #162401, and #162449.
Contribution pattern: security/data-integrity fixes · runtime correctness · agent lifecycle semantics · documentation · schema resolution · updater reliability · multi-agent resource efficiency.
Professional credentials, open-source recognition, and engineering milestones with evidence-linked achievements where available.
CEH — Certified Ethical Hacker, cybersecurity credential associated with ethical hacking, vulnerability assessment, penetration testing, and security fundamentals.
Portfolio status: listed as a professional certification. Credential ID and issue date are intentionally omitted until the original certificate/credential record is published.
Certified AppSec Practitioner (CAP) — The SecOps Group.
Yashas publicly posted that he successfully completed the CAP certification, including a merit/exam announcement. LinkedIn profile evidence.
Contributed a substantial series of upstream OpenClaw fixes spanning reasoning/data integrity, Windows worker transport, agent lifecycle semantics, schema resolution, updater reliability, startup behavior, audit durability, and multi-agent runtime efficiency.
Recognition: maintainer acknowledgements from @steipete and @obviyus are documented in the OpenClaw contributions section.
Built and published engineering work across Pentest-Pi, Marina, AccessDroid-Termux, Termux Coding Agent Toolkit, and browser/MCP agent infrastructure, connecting cybersecurity workflows with mobile AI-agent execution.
Hands-on engineering across VAPT, SOC/SIEM, WAF, cloud security, incident response, security automation, and governance/compliance programs, with tooling experience spanning QRadar, Splunk, Elastic, FortiSIEM/FortiSOAR and major network-security platforms.
Implemented and documented security-management processes covering ISO/IEC 27001 ISMS, access control, incident management, supplier security, business continuity, disaster recovery, privacy, application security, and related audit-ready controls.
Note: this section describes professional compliance work, not a claim of holding an ISO auditor certification.
RajaRajeswari College of Engineering, Bengaluru — 2014–2020.
Compliance focus via ISO/IEC 27001 (ISMS), Annex A controls, and SOC 2/GDPR mapping that threads into VAPT findings as management-system corrective actions.
| Channel | Link |
|---|---|
| linkedin.com/in/yashas-prakash-01a3761a5 | |
| GitHub | github.com/yashas-13 |
| Company | pravidhisolutions.in · LinkedIn company |
| Corporate record | Pravidhi Tech Solutions LLP, LLPIN ACO-8103, Bangalore North — “Prakash Yashas” in corporate records, “Yashas Prakash” on professional profiles. |
For work inquiries: reach out via LinkedIn or Pravidhi Tech Solutions LLP contact. Use the same Yashas Prakash + Bengaluru + yashas-13 identity when searching — multiple homonyms exist.