engage
Router, not method. The driver owns the loop (Skill(offensive), scripts/offensive.py),
the hunt skills own exploitation classes, hunt-core owns discipline. This file exists so no
skill is ever forgotten: every situation below names the exact skill to load. When the
deterministic driver (scripts/offensive.py) is present, its printed next-action wins; when it
is absent, this map IS the router.
Entry preflight (phase 1)
cat targets/active.md names a live engagement (else bash setup/new-engagement.sh <name> <type>).
scope.md filled (in/out-of-scope + RoE) BEFORE any fetch.
python3 scripts/offensive.py index then python3 scripts/offensive-doctor.py
(index-freshness check must pass).
lists/inscope-hosts.md present with at least one row (else run
osint-workflow infra lane or seed from scope.md).
Routing
- Entry: ALWAYS
Skill(workflow-brain) from move 0 (no open-asset
precondition). It locks the first vhost and spawns osint lanes.
- Mid-pass: caido -> js-discover -> js-extractor -> fuzz -> searchsploit/nday
-> hunt-* (signal-gated via dispatch-queue) -> redteamlead (wall or pass
close). The stage table lives in
Skill(active-workflow).
Wiki
Hub: [[methodology-moc]] - Primary page: [[pentest-methodology]]
REFS: [[pentest-methodology]]
1. Pick the driver (it owns the loop)
| Situation |
Skill |
| Any offensive engagement (box/IP foothold->root, bug-bounty *.scope, or client SoW/CIDR/domain) |
Skill(offensive) - offensive.py --type ctf|bb|pentest picks the flavor |
| A standalone challenge (pwn/rev/crypto/forensics/stego/osint/hash/cloud) |
Skill(ctf-category) |
| Driver missing or behaving wrong on this machine |
Skill(offensive)'s "If the driver is unavailable" section + offensive-doctor (scripts/offensive-doctor.py) to diagnose |
| First session on a new machine / VM seat questions |
offensive-doctor (scripts/offensive-doctor.py), then Skill(kali-seat) for the Windows -> WSL -> VM seat |
2. Phase and situation routing (all engagement types)
| Phase / situation |
Skill |
| EVERY new engagement, from move 0 (operator directive 2026-10-06: instant deep-brain; the old 2+ open-rows gate is the floor, not the trigger) |
Skill(workflow-brain) - driver/board/triage stay in the main session (glm-5.3 max effort), wave-1 archetype lanes (fuzz/research/graphify/wiki/redteamlead/hunt) dispatched at once |
| Raw tool output landed (scans, crawls, recon dumps) -> state/loot/Killchain |
Skill(ingest) (cheap-tier subagent; bounded job) |
| "Which tool/payload for this surface" - wiki lookup (quick or deep) |
Skill(wiki-arsenal) |
| External recon / OSINT / attack-surface mapping |
Skill(wiki-recon) |
| Targeted web fuzzing needed (deterministic wordlist pick) |
Skill(fuzz) |
| Driving Burp (proxy triage, Repeater/Intruder/Collaborator) |
Skill(hunt-burp); PoC capture from Burp: Skill(screenshot-burp) |
| Burp/MCP down, per-engagement isolated proxy project wanted, or Caido-native features (Automate, Replay fuzz batteries) |
Skill(caido) (headless Caido on the VM; same load-bearing-visibility discipline, Replay entries instead of Repeater tabs) |
| Manual login / MFA / CAPTCHA the agent cannot do headlessly |
Skill(chrome-devtools-browser) |
| msfconsole work (recon DB, exploits, handler, pivots, post-ex) |
Skill(metasploit) |
| Windows standard-user foothold on a CTF/lab box needs a kernel/LPE shortcut |
Skill(nightmare) (recompiled Windows LPE arsenal with recorded Defender verdicts; authorized CTF/lab only; hub [[privesc-exploit-arsenal]]) |
| Fully-specified exploit compile / escalation run to hand off |
Skill(delegate) (mechanical, cheap tier; false-root guardrail mandatory) |
| Stalled, wrong-vector tell fired, "should I keep hammering?" |
Skill(redteamlead) - NEVER grind; ranked directions + explicit STOP |
| A vuln class never tested on this asset (4a coverage gap) |
Skill(offensive) - offensive.py coverage ranks the gaps |
| Vuln/CVE research on a binary/repo/app/firmware |
Skill(research); ingest a CVE writeup: Skill(nday) / Skill(research-ingest) |
| "Is this valid / should I report it" - validating a finding |
Skill(triage) |
| Evidence hygiene before a report or any share |
Skill(evidence); screenshots: Skill(screenshot) |
| Confirmed/proven finding needs responsible disclosure |
Skill(disclosure) |
| "Why is this row on the board" / which vector baseline applies |
Skill(vector-workflow) |
| Dispatching a lane or verifying what a lane returned |
completion gate (hunt-core "Run completion gate"): python3 scripts/completion-gate.py init <lane-dir> --title "<run>" at dispatch, ... check <lane-dir> --capture <dir> on return; UNCHECKED/MISSING/EMPTY = not done, re-dispatch with the printed nudge |
| Wiki maintenance (re-index, status, page types) |
Skill(wiki) |
Discipline spine for every hunt: Skill(hunt-core) loads alongside any hunt skill (scope gate,
two-account rule, confirmation gate, enumeration limits, stop conditions, FIND output).
3. Scan-mode overlay (quick / standard / deep)
MODE RULES LIVE IN Skill(offensive) + scripts/offensive.py (the mode:
banner) BY REFERENCE - engage routes, it does not restate policy. The
one-line summary: deep is the DEFAULT (operator 2026-10-05), a mode
changes SKIP LIST and PRIORITY ORDER only - never the evidence bar
(a CONFIRMED needs the same working PoC at every mode); shift down
only on operator ask or a status.py budget band; whatever quick skips
is UNEXAMINED, not clean (reopen-if rows in Deadends.md).
4. Fingerprint -> hunt skill
hunt-core's ## Hunt approaches table is the authority - read it; this
index is a memory hook only (class -> skill), never a second source of
routing truth. Experts (xss-, sqli-, ad-*) route through their mappers;
routing.json is the machine-readable copy.
5. Close-out
Run python3 scripts/offensive.py --eng <name> closeout and follow its printed order - the chain
lives only in CLOSEOUT_CHAINS (scripts/offensive.py), not restated here to avoid drift.
Maintenance rule
A skill added to skills/ without a routing row in this file or in scripts/routing.json is an
orphan by definition: nothing will ever name it mid-engagement. New skill -> add its row to the
section above that fits.
Entry gate
Router-scaled, falsifiable: (1) targets/active.md names an engagement
(or route to scaffolding); (2) the routing table you rely on is FRESH -
python3 scripts/offensive.py --eng <eng> index exits 0 and the
.offensive-index.json mtime postdates the last routing.json edit (a
stale index routes by dead rows); (3) scope.md parses (frontmatter +
in-scope list). A failed check routes to the fix, not to a hunt lane.
1---2name: engage3description: "Orchestrator and skill router: the single situation-to-skill map across the whole library (driver, hunt-*, process skills). Use at session start, whenever between phases, or unsure which skill is next."4metadata:5 source: https://skillmd.com/skills/encod3d-sec/engage6 hermes:7 tags: [router, orchestration, triage]8 related_skills: [offensive, brain, hunt-core]9---1011# engage1213Router, not method. The driver owns the loop (`Skill(offensive)`, `scripts/offensive.py`),14the hunt skills own exploitation classes, `hunt-core` owns discipline. This file exists so no15skill is ever forgotten: every situation below names the exact skill to load. When the16deterministic driver (`scripts/offensive.py`) is present, its printed next-action wins; when it17is absent, this map IS the router.1819## Entry preflight (phase 1)20211. `cat targets/active.md` names a live engagement (else `bash22 setup/new-engagement.sh <name> <type>`).232. `scope.md` filled (in/out-of-scope + RoE) BEFORE any fetch.243. `python3 scripts/offensive.py index` then `python3 scripts/offensive-doctor.py`25 (index-freshness check must pass).264. `lists/inscope-hosts.md` present with at least one row (else run27 osint-workflow infra lane or seed from scope.md).2829## Routing3031- Entry: ALWAYS `Skill(workflow-brain)` from move 0 (no open-asset32 precondition). It locks the first vhost and spawns osint lanes.33- Mid-pass: caido -> js-discover -> js-extractor -> fuzz -> searchsploit/nday34 -> hunt-* (signal-gated via dispatch-queue) -> redteamlead (wall or pass35 close). The stage table lives in `Skill(active-workflow)`.3637## Wiki3839Hub: [[methodology-moc]] - Primary page: [[pentest-methodology]]40**REFS:** [[pentest-methodology]]4142## 1. Pick the driver (it owns the loop)4344| Situation | Skill |45|---|---|46| Any offensive engagement (box/IP foothold->root, bug-bounty *.scope, or client SoW/CIDR/domain) | `Skill(offensive)` - `offensive.py --type ctf\|bb\|pentest` picks the flavor |47| A standalone challenge (pwn/rev/crypto/forensics/stego/osint/hash/cloud) | `Skill(ctf-category)` |48| Driver missing or behaving wrong on this machine | `Skill(offensive)`'s "If the driver is unavailable" section + `offensive-doctor` (`scripts/offensive-doctor.py`) to diagnose |49| First session on a new machine / VM seat questions | `offensive-doctor` (`scripts/offensive-doctor.py`), then `Skill(kali-seat)` for the Windows -> WSL -> VM seat |5051## 2. Phase and situation routing (all engagement types)5253| Phase / situation | Skill |54|---|---|55| EVERY new engagement, from move 0 (operator directive 2026-10-06: instant deep-brain; the old 2+ open-rows gate is the floor, not the trigger) | `Skill(workflow-brain)` - driver/board/triage stay in the main session (glm-5.3 max effort), wave-1 archetype lanes (fuzz/research/graphify/wiki/redteamlead/hunt) dispatched at once |56| Raw tool output landed (scans, crawls, recon dumps) -> state/loot/Killchain | `Skill(ingest)` (cheap-tier subagent; bounded job) |57| "Which tool/payload for this surface" - wiki lookup (quick or deep) | `Skill(wiki-arsenal)` |58| External recon / OSINT / attack-surface mapping | `Skill(wiki-recon)` |59| Targeted web fuzzing needed (deterministic wordlist pick) | `Skill(fuzz)` |60| Driving Burp (proxy triage, Repeater/Intruder/Collaborator) | `Skill(hunt-burp)`; PoC capture from Burp: `Skill(screenshot-burp)` |61| Burp/MCP down, per-engagement isolated proxy project wanted, or Caido-native features (Automate, Replay fuzz batteries) | `Skill(caido)` (headless Caido on the VM; same load-bearing-visibility discipline, Replay entries instead of Repeater tabs) |62| Manual login / MFA / CAPTCHA the agent cannot do headlessly | `Skill(chrome-devtools-browser)` |63| msfconsole work (recon DB, exploits, handler, pivots, post-ex) | `Skill(metasploit)` |64| Windows standard-user foothold on a CTF/lab box needs a kernel/LPE shortcut | `Skill(nightmare)` (recompiled Windows LPE arsenal with recorded Defender verdicts; authorized CTF/lab only; hub [[privesc-exploit-arsenal]]) |65| Fully-specified exploit compile / escalation run to hand off | `Skill(delegate)` (mechanical, cheap tier; false-root guardrail mandatory) |66| Stalled, wrong-vector tell fired, "should I keep hammering?" | `Skill(redteamlead)` - NEVER grind; ranked directions + explicit STOP |67| A vuln class never tested on this asset (4a coverage gap) | `Skill(offensive)` - `offensive.py coverage` ranks the gaps |68| Vuln/CVE research on a binary/repo/app/firmware | `Skill(research)`; ingest a CVE writeup: `Skill(nday)` / `Skill(research-ingest)` |69| "Is this valid / should I report it" - validating a finding | `Skill(triage)` |70| Evidence hygiene before a report or any share | `Skill(evidence)`; screenshots: `Skill(screenshot)` |71| Confirmed/proven finding needs responsible disclosure | `Skill(disclosure)` |72| "Why is this row on the board" / which vector baseline applies | `Skill(vector-workflow)` |73| Dispatching a lane or verifying what a lane returned | completion gate (hunt-core "Run completion gate"): `python3 scripts/completion-gate.py init <lane-dir> --title "<run>"` at dispatch, `... check <lane-dir> --capture <dir>` on return; UNCHECKED/MISSING/EMPTY = not done, re-dispatch with the printed nudge |74| Wiki maintenance (re-index, status, page types) | `Skill(wiki)` |7576Discipline spine for every hunt: `Skill(hunt-core)` loads alongside any hunt skill (scope gate,77two-account rule, confirmation gate, enumeration limits, stop conditions, FIND output).7879## 3. Scan-mode overlay (quick / standard / deep)8081MODE RULES LIVE IN Skill(offensive) + scripts/offensive.py (the `mode:`82banner) BY REFERENCE - engage routes, it does not restate policy. The83one-line summary: deep is the DEFAULT (operator 2026-10-05), a mode84changes SKIP LIST and PRIORITY ORDER only - never the evidence bar85(a CONFIRMED needs the same working PoC at every mode); shift down86only on operator ask or a status.py budget band; whatever quick skips87is UNEXAMINED, not clean (reopen-if rows in Deadends.md).8889## 4. Fingerprint -> hunt skill9091`hunt-core`'s `## Hunt approaches` table is the authority - read it; this92index is a memory hook only (class -> skill), never a second source of93routing truth. Experts (xss-*, sqli-*, ad-*) route through their mappers;94routing.json is the machine-readable copy.9596## 5. Close-out9798Run `python3 scripts/offensive.py --eng <name> closeout` and follow its printed order - the chain99lives only in `CLOSEOUT_CHAINS` (`scripts/offensive.py`), not restated here to avoid drift.100101## Maintenance rule102103A skill added to `skills/` without a routing row in this file or in `scripts/routing.json` is an104orphan by definition: nothing will ever name it mid-engagement. New skill -> add its row to the105section above that fits.106107## Entry gate108109Router-scaled, falsifiable: (1) targets/active.md names an engagement110(or route to scaffolding); (2) the routing table you rely on is FRESH -111`python3 scripts/offensive.py --eng <eng> index` exits 0 and the112.offensive-index.json mtime postdates the last routing.json edit (a113stale index routes by dead rows); (3) scope.md parses (frontmatter +114in-scope list). A failed check routes to the fix, not to a hunt lane.