pentest-workflow

name	pentest-workflow
description	Orchestrates PTES-aligned phases, gates, and subgates for pentest engagements. Each subgate links to a skill the agent must load and execute. Use when starting an engagement, advancing phases, completing subgates, or when the user asks about workflow, gates, checkpoints, or pentest methodology structure.

Pentest Workflow: Phases, Gates, Subgates

Architecture

Phase (PTES macro)
  └── Gate (exit criteria before next phase)
        └── Subgates (atomic tasks)
              └── Skill (agent loads and executes)
                    └── Agent (optional delegate)
                          └── Skill content (self-contained techniques)

Framework synthesis:

• PTES defines 7 macro phases (state machine)
• NIST SP 800-115 allows loopback: post_exploit -> recon when new scope found
• OWASP WSTG provides web analyze/recon subgates (web_app type)
• MITRE ATT&CK tags internal post-exploit subgates (internal_ad type)
• Unified Kill Chain tracks objective depth (In / Through / Out)

Full research: frameworks.md

Configuration files

File	Purpose
engagements/catalogs/phases.yaml	Phase definitions, gates, skill/agent routing
engagements/catalogs/subgates-common.yaml	Pre-engage, model, exploit, report subgates
engagements/catalogs/subgates-web-app.yaml	OWASP WSTG subgates
engagements/catalogs/subgates-internal-ad.yaml	ATT&CK subgates
engagements/{id}/phase-state.yaml	Current phase, completed subgates
engagements/{id}/world-state.md	Human-readable action log + gate/subgate status
engagements/{id}/world-state-index.json	Machine index for duplicate-action detection
engagements/{id}/objectives.yaml	Engagement objectives with status
engagements/{id}/targets.yaml	Prioritized targets (model phase)
engagements/{id}/roe.yaml	ROE + engagement_type

Phase overview

Phase	PTES	Gate	Primary agent	Core skills
pre_engage	1	G0 (hard)	harness-orchestrator	msf-harness
recon	2	G1	msf-recon-agent	msf-recon, hacktricks-methodology
model	3	G2 (hard)	harness-orchestrator	pentest-knowledge-base
analyze	4	G3	msf-recon-agent	web-app-pentest OR internal-ad-pentest
exploit	5	G4 (hard)	msf-exploit-agent	msf-exploit-chain
post_exploit	6	G5	msf-post-agent	msf-post, internal-ad-pentest
report	7	G6 (hard)	harness-orchestrator	pentest-knowledge-base

Agent operating procedure

1. Read state

engagements/{engagement_id}/world-state.md
engagements/{engagement_id}/phase-state.yaml
engagements/{engagement_id}/roe.yaml
engagements/catalogs/phases.yaml

Initialize or refresh world state when starting work:

MSF MCP (preferred): No direct module. Use workflow scripts below.

CLI fallback:

python .cursor/skills/pentest-workflow/scripts/world-state.py init <engagement_id>
python .cursor/skills/pentest-workflow/scripts/world-state.py sync <engagement_id>

Hooks on shell and MCP calls will:

• Before execution: warn the agent if the same action fingerprint already exists
• After execution: append command/tool output to world-state.md

2. Check gate and pending subgates

MSF MCP (preferred): No direct module. Use gate-check script.

CLI fallback:

python .cursor/skills/pentest-workflow/scripts/gate-check.py <engagement_id>

3. RPC connectivity subgate (SG0.2)

MSF MCP (preferred):

msf_status()

CLI fallback:

wsl -e bash -lc "msfconsole -q -x 'version; exit'"

Mark complete:

python .cursor/skills/pentest-workflow/scripts/complete-subgate.py SG0.2-rpc-check --engagement <engagement_id> --evidence "msf_status: connected"

4. Work subgates in order

For each PENDING subgate:

1. Load the subgate's skill (from catalog YAML)
2. Execute the subgate tasks using that skill's workflow
3. Verify exit_criteria
4. Mark complete:

MSF MCP (preferred): No direct module. Use complete-subgate script.

CLI fallback:

python .cursor/skills/pentest-workflow/scripts/complete-subgate.py SG0.2-rpc-check --engagement <engagement_id> --evidence "msf_status: connected"

5. Save evidence to evidence/msf/ with subgate ID in filename when possible

5. Pre-exploit gates (SG4.1 and SG4.3 are SEPARATE subgates)

SG4.1: Reviewer Gate (approval by /msf-reviewer)

This is a security REVIEW, not a technical check. The /msf-reviewer agent reads scope, ROE, and proposed action, then produces an APPROVED/DENIED/NEEDS_INFO verdict.

Completion evidence: evidence/msf/review-{engagement_id}-{date}.md with verdict APPROVED.

SG4.3: Module Vulnerability Check (technical probe)

This is the actual msf_module_check call that probes the target:

msf_module_check(
  engagement_id="<id>",
  module_type="exploit",
  module_name="<module>",
  options={"RHOSTS": "<target>"}
)

Parse the check_result:

• "vulnerable" -> proceed to exploit
• "safe" / "not vulnerable" -> STOP, do not exploit
• "cannot be determined" / "not supported" -> proceed with caution

Order: SG4.1 MUST complete before SG4.3. SG4.3 MUST complete before exploit.

Reviewer gate must pass before any msf_run_exploit call.

6. Advance phase

When all subgates pass and gate criteria met:

MSF MCP (preferred): No direct module. Use gate-check with advance flag.

CLI fallback:

python .cursor/skills/pentest-workflow/scripts/gate-check.py <engagement_id> --advance

Or update phase-state.yaml manually with phase_history entry.

7. Delegate to subagents

When subgate specifies an agent, delegate:

Subgate agent	When
/harness-orchestrator	Phase transitions, model, report
/msf-recon-agent	Recon and analyze subgates
/msf-exploit-agent	Exploit subgates (SG4.*)
/msf-post-agent	Post-exploit subgates
/msf-reviewer	SG4.1 reviewer gate (mandatory before exploit)

Engagement types

Type	Recon/analyze catalog	Post catalog
web_app	subgates-web-app.yaml	minimal post
internal_ad	subgates-internal-ad.yaml	full ATT&CK through
hybrid	both (web first, then AD)	full
external	internal_ad recon + hacktricks	scope-limited

Set engagement_type in roe.yaml.

Loopback (NIST 800-115)

From post_exploit, if subgate SG-I-LOOPBACK finds new hosts/services:

1. Increment loopback_count in phase-state.yaml
2. Set current_phase: recon
3. Add note to phase_history with reason
4. Re-run recon subgates for new scope only

Hard gates (never skip)

Gate	Requirement
G0	Scope + ROE + RPC
G2	targets.yaml + objectives.yaml
G4	Reviewer approval + msf_module_check
G6	Findings documented + sessions cleaned

Subgate -> skill map (quick reference)

Subgate prefix	Skill
SG0.*	msf-harness, pentest-workflow
SG-W-*	web-app-pentest
SG-I-RECON*, SG-I-AD*	hacktricks-methodology, internal-ad-pentest
SG-I-VULN*, SG-I-CRED-PATHS	msf-recon, internal-ad-pentest
SG2.*	pentest-knowledge-base, methodology-cheatsheets
SG4.*	msf-exploit-chain
SG-I-DISCOVERY..SG-I-DOMAIN	internal-ad-pentest, msf-post
SG-I-ADCS, SG-I-CVE	internal-ad-pentest
SG-I-PERSIST, SG-I-PIVOT, SG-I-EVASION	internal-ad-pentest
SG-I-CLOUD	cloud-pentest
SG-I-CONTAINERS	container-devops-pentest
SG-I-LINUX	linux-pentest
SG-I-MACOS	macos-pentest
SG-W-AI	ai-llm-pentest
SG6.*	pentest-knowledge-base, msf-post

Output format (orchestrator)

When reporting workflow progress:

## Workflow Status

**Phase:** recon (Intelligence Gathering)
**Gate:** G1-recon-complete - BLOCKED

### Subgates
- [x] SG0.1-scope-verify (msf-harness)
- [ ] SG-I-RECON-PASSIVE (hacktricks-methodology) <- CURRENT

### Next action
Load hacktricks-methodology skill, run msf_host_info + service enum, save evidence/msf/recon-{date}.txt, then complete-subgate.

Related skills

• pentest-knowledge-base - skill routing
• web-app-pentest - OWASP/WSTG execution
• internal-ad-pentest - AD/ATT&CK execution
• hacktricks-methodology - Service pentest
• cloud-pentest - AWS/Azure/IBM cloud abuse
• container-devops-pentest - Docker/K8s/CI-CD
• mobile-pentest - Android/iOS when in ROE
• macos-pentest - macOS privesc and red teaming
• linux-pentest - Linux privesc and post-ex
• ai-llm-pentest - LLM prompt injection and AI MCP abuse
• binary-exploit-pentest - Binary exploitation when ROE allows
• methodology-cheatsheets - methodology and tool refs
• msf-recon, msf-exploit-chain, msf-post - MCP tool workflows
• msf-harness - Setup and safety