# Exploit Specialist Methods

> Evidence-led exploit technical methods and evidence interpretation for the relevant environment.

- Skill: `purpleailab/exploit-specialist-methods` (Agent Skill)
- Install (CLI): `npx get-skill add purpleailab/exploit-specialist-methods`
- Raw SKILL.md: https://api.skillmd.com/api/skills/purpleailab/exploit-specialist-methods/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: purpleailab (https://skillmd.com/u/purpleailab)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/purpleailab/exploit-specialist-methods

---


# Technical methods and evidence

Use the relevant existing technique skills for implementation details. These notes
explain technical prerequisites and evidentiary limits; they do not define an
engagement stage order, dispatch policy, or report lifecycle. Tool availability and
argument names come from the live schema. Authorization comes from the RoE.

## Reproducer design

Bind a proof to exact target/version, required identity, request bytes, session state,
input and expected observable. A new GET may rotate CSRF tokens, cookies or encrypted
challenges; do not submit data recovered under session A using session B. Preserve
runnable scripts and redacted configuration with raw response artifacts. Copy a
running harness before editing it and retain resume state for long extractions.

## Component and protocol fidelity

Version-specific behavior can depend on proxy tiers, HTTP framing, runtime patch
levels or exact gadget classes. Use available `cve_lookup`, `cve_poc_lookup`,
`methodology_lookup` and `payload_search` when those facts are needed; an advisory
match does not prove local exploitability. Respect byte-level prerequisites rather
than substituting a superficially similar payload. Host-header routing, internal
references and frontend/backend differences identify hypotheses, not new scope.

## Discriminating observables

- Browser execution requires a browser-observable effect and realistic delivery;
  HTTP reflection alone is insufficient. Use an available browser tool only when
  the proof depends on JS, navigation or browser security behavior.
- Cross-account access requires controlled identities, known object ownership and
  a content or state comparison; status-code differences alone are ambiguous.
- Injection requires an output or differential tied to the injected condition,
  with syntax/encoding controls that rule out ordinary error handling.
- In-band SSRF may be established by attributed response content; blind SSRF needs
  a correlated owned-listener interaction. OOB is not mandatory for in-band cases.
- Races and timing channels need repeated matched trials with fresh state, baselines
  and noise characterization. A single fast response or race win is inconclusive.
- Request desynchronization requires observable cross-request framing/state effects;
  differential 400/501 responses alone are not confirmation.

Use existing domain routers and technique skills for exact test construction.
Network harnesses need inner I/O timeouts and an outer process deadline; follow
injected bash/session semantics and terminate only identified owned processes.

