RedScope AI is a security-engineering and authorized red-team assistant. Use it only against systems, networks, accounts, logs, and repositories you are explicitly authorized to assess.

• Written rules of engagement (RoE) / scope documents are required for active testing.
• Scope, rate limits, and evidence retention are first-class project constraints — not optional.
• The project does not support unauthorized scanning, mass targeting, or detection evasion for malicious purposes.

On first launch RedScope can prompt (or you can run bun run redscope:first-run) to:

1. Scrape ~500 public free-proxy endpoints from public list pages, for IP rotation during authorized testing. The pool auto-refreshes about every 6 hours when you opted in.
2. Silently collect ≥100 PoC / n-day references (CVE metadata + official advisory links only). Individual entries are not dumped to the console.

Both steps are optional. Declining is fine; you can run the collectors later:

bun run redscope:first-run
bun run redscope:proxy-scrape -- --limit 500
bun run redscope:poc-catalog

Public free-proxy nodes are noisy and short-lived. Always health-check before use, and only send traffic at in-scope targets.

• Help plan and execute authorized penetration tests, recon, and analysis inside a declared scope.
• Auto-write per-host progress / activity / findings into redscope-engagement.json and refresh the recon map during testing.
• Index public vulnerability / advisory metadata and, when a target fingerprint matches, capture related n-day references into the local catalog.
• Rotate among user-opted public free proxies and/or referee-provided egress nodes.
• Gate “vulnerability confirmed” behind evidence-based verification — version banners alone are not enough (high false-positive risk).

• Claim a host is vulnerable from a version string / Server banner alone.
• Ship or generate weaponized exploit payloads inside this repository. RedScope stores CVE/advisory references, fingerprint→n-day metadata, and evidence-gated verification hooks. Live exploitation tooling (if any) must come from external, operator-installed, authorized scanners (e.g. nuclei under an approved profile) and still pass the evidence gate before a finding is marked confirmed/exploited.
• Bypass the permission / auto-mode gates without an explicit full-access (bypassPermissions / auto) opt-in from the operator.
• Test out-of-scope targets. Full-access autonomy still requires a declared engagement scope.

While a test step runs, RedScope can:

1. Health-check the current egress IP (TCP)
2. On failure / HTTP block (403/407/429/…) silently cool that node
3. Switch to the next live proxy and continue the same request/step

import { withSilentEgress, fetchWithSilentEgress } from './src/utils/silentEgress.ts'
import { startTestStep, probeFingerprintAndCaptureNdays, finishTestStep } from './src/utils/testSession.ts'

No operator prompt mid-test — rotation is automatic within the configured pool.

postinstall runs scripts/postinstall-redscope-bootstrap.cjs, which (unless skipped) downloads ~500 public free-proxy nodes and builds ≥100 PoC references into the user config / workspace. Non-fatal on failure.

# skip
REDSCOPE_SKIP_FIRST_RUN_BOOTSTRAP=1 npm i -g @redscope-ai/redscope
# force redo
REDSCOPE_FORCE_FIRST_RUN_BOOTSTRAP=1 bun run scripts/postinstall-redscope-bootstrap.cjs

# Public free-proxy pool (after first-run opt-in or manual scrape)
bun run redscope:proxy-scrape -- --limit 500
bun run redscope:proxy-scrape -- --status

# Health-check + select through whatever pool loadEgressConfig() resolves
bun run redscope:egress-refresh -- --status
bun run redscope:egress-refresh -- --check
bun run redscope:egress-refresh -- --select --target <in-scope-host>

loadEgressConfig() preference order:

1. REDSCOPE_EGRESS_CONFIG / REDSCOPE_TOOLS_EGRESS_CONFIG
2. Public free-proxy pool (public-free-proxies.json) when present
3. Referee-provided authorized egress config

bun run redscope:fingerprint-nday -- --url https://in-scope.example
bun run redscope:fingerprint-nday -- --url https://in-scope.example --host-id ext-web

Matching products pull local catalog hits and (unless --offline) query NVD keyword search. New CVE references are appended to redscope-poc-catalog.json for next time. This stores metadata + links, not exploit code.

When template verification is needed and nuclei is not installed, RedScope prompts the operator to download it into the current project:

bun run redscope:nuclei-setup
# → tools/bin/nuclei(.exe)

After install, maybeRunNucleiVerification() / profile steps can invoke it automatically against in-scope targets. Hits still pass the evidence gate before a finding is confirmed.

pocVerification.judgeVerification() refuses to confirm findings when the only signal is a version/banner match. Confirmation requires non-version evidence (body marker, file-read contents, auth-bypass proof, OOB token, scanner template hit with proof, etc.). Use verifyAndRecordFinding() to write through to the engagement graph with progress auto-update.

import {
  beginHostTest,
  reportHostProgress,
  reportHostFinding,
  completeHostTest,
} from './src/utils/engagementProgress.ts'

These helpers persist redscope-engagement.json and regenerate redscope-recon-map.html so the HUD and SVG map stay live during a test.

bun run redscope:poc-catalog

Writes / refreshes redscope-poc-catalog.json (≥100 scope-gated public CVE/advisory references). Planning material, not an executable arsenal.

bun run redscope:recon-map

If you find a security issue in this project (the CLI, build pipeline, or bundled scripts), please open a private report via the repository’s security advisory channel or contact the maintainers.