| name | planning |
| description | Pre-execution planning and priority-scored task list before running tools. |
planning
Objective
Create a clear, low-noise attack plan before any active testing.
Mandatory pre-checks
- Read research-notes for current high-signal bug classes.
- Use traffic-proxy skill only after a concrete workflow hypothesis exists.
- Scope validation
- Confirm in-scope domains/IPs.
- Confirm out-of-scope exclusions.
- Risk tiering
- Public internet production targets: low-noise only first.
- Internal labs/CTF: broader active scans allowed.
- Initial target profile
- Domain, ASN, technology hints, auth surface, likely API presence.
Priority scoring model
Score each task by impact x confidence x feasibility (1-5 each).
- High score (60+): execute first.
- Medium (30-59): execute after first evidence.
- Low (<30): defer.
Output contract before scanning
Produce TODO list with:
- Phase: recon / enum / vuln / verify / report
- Tool: exact tool name
- Purpose: one sentence
- Success signal: what output means progress
- Stop condition: when to stop this task
Example TODO item
- Phase: recon
- Tool: subfinder
- Purpose: collect passive subdomains for attack surface
- Success signal: non-empty unique subdomain set
- Stop condition: duplicate-only output in 2 consecutive runs
Execution rule
Never run exploitation until at least one strong vulnerability signal is confirmed in recon/enum.
2024-2026 high-signal focus
Prioritize bug classes that repeatedly show high bounty impact:
- Broken access control (IDOR/BOLA/role boundary failures)
- Authentication/session logic flaws
- API data exposure and object-level authorization issues
- Misconfiguration and sensitive endpoint exposure
Endpoint-first planning
Before heavy scans, build a shortlist of high-value endpoint clusters:
- Account and profile endpoints
- Payment and transaction workflows
- Admin and internal API routes
- File upload and export/import endpoints
For each cluster, define:
- Expected actor (anonymous/user/admin/service)
- Expected object boundary (self-only/team/global)
- One negative test case (forbidden access)
MCP reliability
- For long-running tools, use async flow: start_tool -> job_status -> job_result.
- Keep max 2 concurrent jobs and 1 heavy scan per target in low-noise mode.