Your first engagement

We’ll run a real AEGIS engagement against hackthebox.com’s public CTF lab so you can see every phase fire without breaking any laws. Always confirm authorisation before running AEGIS against any target.

1. Create the engagement

aegis engagement new \
  --client "Personal lab" \
  --scope juice-shop.herokuapp.com \
  --budget-usd 2.00

A new folder engagements/<id>-personal-lab/ appears with:

engagements/<id>-personal-lab/
├── scope.yaml          # what's in/out of scope, rate limits, exclusions
├── aegis.db            # SQLite — findings, audit log, tool runs
└── artifacts/          # raw tool output, per-tool subfolders

2. Inspect the scope

Open scope.yaml:

engagement_id: PERSONAL-LAB-2026-001
client: Personal lab
in_scope:
  - juice-shop.herokuapp.com
out_of_scope: []
rules:
  rate_limit_rps: 5
  max_concurrent_tools: 4
  authorisation: confirmed
  expires_at: 2026-12-31

Edit rate_limit_rps if you want to be gentler. Anything outside in_scope is hard-rejected.

3. Run it

aegis run engagements/<id>-personal-lab

The TUI takes over. Watch:

• Status bar — phase, cost, tokens, severity counts.
• Activity feed (left) — every tool the agent fires as a card with the command and (in expanded mode, Ctrl+O) its output.
• Agent pane (right) — what Claude is reasoning about right now, in italics.

A typical first run takes 15–40 minutes and costs $0.50–$3.00 in Anthropic tokens.

4. Steer the agent mid-flight

You can talk to the agent while it works. Inside the TUI:

> check if the rest api at /api/v3 enforces csrf

That’s queued as an operator message and injected into the next phase’s prompt. Claude will respond in its aegis_note events (visible in the right pane).

Slash commands:

Command	Effect
/stop	Wrap up after the current phase
/pause / /resume	Suspend / continue
/focus <target>	Tell the agent to prioritise a target
/hunt on	After REPORTING, loop back to ACTIVE_RECON for another pass
/quit	Exit immediately (abandons the scan)

5. Read the findings

When AEGIS finishes, it prints a one-page summary then drops you back in the shell. Generate the full report:

aegis report engagements/<id>-personal-lab

This makes report.html, report.md, and report.json. The JSON is machine-readable; pipe it to Burp, JIRA, or your bug-bounty platform of choice.

6. Re-run targeted phases

Already ran the full thing but want to dig into one phase?

aegis run engagements/<id>-personal-lab --phase vuln_analysis

This skips straight to VULN_ANALYSIS and uses everything already in the engagement DB (subdomains, endpoints, fingerprints) as starting context.

What you just saw

• Real MCP tool calls. Claude didn’t write a single line of code; it called typed MCP tools that AEGIS surfaced. Every call passed the scope guard, rate limiter, and audit log.
• Adaptive escalation. Findings in one phase queue follow-ups in the next. A CVE in nuclei → nuclei-fuzz. A leaked secret → trufflehog deeper. A weird redirect → OAuth probe.
• Verification before reporting. Candidates went through 33 verification probes before they made it into the report. No hallucinated SQLi.

Next: CLI reference — every command, every flag.