Techvify Forge Secure · Security

Penetration testing, run by agents.

Web · mobile · API · network · graybox — durably orchestrated.

Forge Secure is an AI penetration-testing framework. A team of specialist agents discovers the attack surface, maps every role, hunts five vulnerability axes in parallel, and proves each finding with a real exploit and recorded evidence — then writes the report.

Authorized targets only.

5
Pipeline tiers
13–15
AI agents
5
Vuln axes
forge secureweb grayboxrun #318live

▸ web graybox engagement · target: acme

  • passed: reconsurface mapped · 47 routesbrowser
  • passed: authmulti-persona session capturedadmin · staff
  • finding: injSQLi · /api/searchcritical
  • finding: authzIDOR · patient recordscritical
  • warning: xssstored XSS · bio fieldhigh
  • passed: exploit7 findings reproducedtraces ✓
  • passed: reportfinal_report · DOCX + PDFdone

▮ deliverables

  • injection_analysis.md
  • authz_evidence.md
  • xss_queue.json
  • traces/admin/….zip
  • final_report.md
  • final_report.docx
2
critical
3
high
5
medium
  1. Recon
  2. Auth
  3. Analyse
  4. Exploit
  5. Report

What's inside

Exploits, not alerts.

Five pipeline tiers, one CLI

Web, mobile, API, internal network, and source-aware whitebox — each shares the same discover → analyze → exploit → report skeleton, with the right tooling wired in automatically.

A team of 13–15 AI agents

Discovery, auth-mapper, five parallel vuln specialists, five pipelined exploit agents, and a report writer. Each has one job and explicit, scoped tool access.

Exploits, not just alerts

Every finding is confirmed with a real payload and a recorded browser-automation trace — reproducible evidence, not scanner-grade false positives.

Multi-persona access control

Tests cross-role IDOR and privilege escalation by reading every persona’s token bundle at once — the class of bug a single-account scan literally cannot find.

Durable & resumable

A durable workflow engine orchestrates the whole run. Agents are git-checkpointed and retried automatically; a paused or interrupted engagement resumes exactly where it left off.

Branded reports + re-test

Assembles an executive report (DOCX + PDF) on Techvify letterhead, and re-verifies prior remediation claims so a re-test round closes the loop.

Prerequisites.

Forge Secure runs as a full engagement framework. The agent plugin wraps a CLI that needs Docker (it brings up a durable workflow engine plus a worker container) and its own model provider credential — that credential, not your agent subscription, pays for the pentest agents.

Five pipeline tiers

One CLI. Five kinds of target.

Each tier shares the same discover → analyze → exploit → report skeleton, with the right tooling wired in automatically.

  • Web graybox

    A web app with no source. The agents drive a real browser — browser-driven DAST.

    13 agents · browser automation

  • Mobile graybox

    Native iOS / Android. Device automation drives a real device or emulator while the agents probe the live UI.

    13 agents · device automation

  • API graybox

    A pure HTTP API, no browser. Schema-driven from OpenAPI / GraphQL.

    13 agents · schema-driven

  • Network graybox

    Internal network / Active Directory. Discovery, exploitation and attack-path tooling — adds enumeration + post-exploit phases.

    15 agents · network tooling

  • Whitebox web

    You have source access. Agents read the code to guide attacks, then validate against the running web app.

    13 agents · source + DAST

The pipeline

Discover, then exploit — pipelined.

Discovery and auth-mapping run once. Five vuln agents fan out in parallel; each exploit agent starts the instant its vuln agent finishes — no barrier — so the XSS exploit can run while injection is still analysing. Every run is durable and resumable.

Tier 0

Discovery / reconnaissance

Maps the attack surface — routes, forms, endpoints, tech stack. Web is crawled with browser automation; API ingests OpenAPI / GraphQL schemas; network runs discovery and a sequential enumeration phase first.

  • discovery
  • enumeration

Watch an engagement end to end

The recorded run — surface mapped, findings reproduced with real exploits, report written — is open to Techvify users and invited guests.

Coverage

Five vulnerability axes, exploited and proven.

Aligned to the OWASP Web Security Testing Guide. Every finding is confirmed with a real payload and a recorded trace — no scanner-grade false positives.

  • InjectionSQLi, command and template injection — probed on every parameter, header, and form.
  • XSSReflected, stored, and DOM-based cross-site scripting, confirmed with live payloads.
  • Broken authWeak session handling, login bypasses, token and 2FA flaws.
  • SSRFServer-side request forgery against internal services and metadata endpoints.
  • Access controlIDOR, vertical and horizontal privilege escalation — cross-persona by construction.

Out of scope: vulnerable dependencies, weak crypto, and configuration hardening are deferred to static analysis, not this dynamic engine. Cloud-asset testing is a separate future tier.

Beyond a single scan

Multi-persona by design. Remediation re-tested.

Multi-persona testing

Cross-role IDOR + privilege escalation

Discovery runs once; auth-mapper and the vuln/exploit pairs run per persona in parallel. The authz agent reads every persona's token bundle to test boundaries a single-account scan cannot reach. Two personas cost ~1.5–2× a single run.

Remediation verification

Re-test what DevOps claims to have fixed

Drop prior fix claims into the remediation folder and the report agent re-tests each one — rendering a verdict into the final report. No folder means it behaves as a first-round assessment.

  • Verified Fixed
  • Still Vulnerable
  • Partially Fixed
  • Inconclusive

The roster

A team of specialist agents, one per job.

13 agents for web / mobile / API / whitebox, 15 for network. Each is a prompt with explicit, scoped tool access — durably orchestrated, with automatic retries and git-checkpointed state.

discovery
Maps the surface — crawls the app or ingests API schemas, fingerprints the stack, and hands a route/endpoint inventory to everyone downstream.
enumerationnetwork only
Discovery, exploitation tooling and service enumeration build the host/service graph before vuln analysis begins.
auth-mapper
Logs in per persona, captures session tokens, and produces the role + permission matrix the authz agent reads for cross-role tests.
vuln-injection
Hunts SQL, command, template injection and path traversal across parameters, headers, and bodies — writes an analysis deliverable + exploit queue.
vuln-xss
Finds reflected, stored, and DOM cross-site scripting; queues the payloads that actually fire.
vuln-auth
Tests authentication — login bypass, weak sessions, credential and token handling.
vuln-ssrf
Probes for server-side request forgery, including reach to internal services and cloud-metadata endpoints.
vuln-authz
Reads every persona’s token bundle to find IDOR, BFLA, and vertical/horizontal privilege escalation — the class single-persona tools cannot find.
exploit-injection
Reproduces queued injection candidates with a real payload and saves a replayable trace as evidence.
exploit-xss
Fires queued XSS payloads to prove execution in the browser; records the trace.
exploit-auth
Demonstrates auth bypass or session takeover for queued authentication candidates.
exploit-ssrf
Proves SSRF reach against internal services or metadata, with recorded evidence.
exploit-authz
Confirms cross-role access — IDOR and privilege escalation — by replaying as a lower-privilege persona.
report
Re-tests remediation claims, correlates findings, and renders the branded executive report (DOCX + PDF) with an access-control matrix.

Every command, what it does.

The full CLI reference — starting a run per tier, tailing logs, opening the evidence trace, exporting the report, and resuming a paused engagement — lives in the guidebook.

Open the CLI reference

Setup steps live in the guidebook.

The guidebook carries the current install and configuration steps for every kit, kept in one place so they stay accurate. Open to Techvify users and invited guests.

Open the guidebook

Start

Gate your next release.

Bring one application and its next release. On a scoping call we agree what each dimension checks, what fails the run and who signs, before any work starts.