Distilled from reviewing real AI-pentest output that kept stopping at "exposed"
instead of "exploited". Pure-additive, back-compatible.
Behavior (injected into black/grey/chain exploit prompts via DEPTH_DOCTRINE):
- Exposed → exploited: any info-disclosure / exposed service/WSDL / leaked
credential|token / reachable dev host MUST be used before it's a finding;
otherwise it's a lead, not a confirmed High/Critical.
- Chain across modules: reuse obtained session/JWT/cookie/credential and pivot
to IDOR/privesc/exfil; report the chain, not isolated parts.
- Decode & fingerprint → CVE; audit tokens (alg-confusion/none/kid/JWKS, weak
HS256 secret cracking, lifecycle).
Deterministic post-pass (new crates/harness/src/hygiene.rs, wired into finish()):
- calibrate severity to PROVEN impact — unproven High/Critical (hedged, no
payload, thin evidence) capped to Medium and re-titled "(potential)";
- depth_audit — flag exposures on a host with no real exploit;
- hygiene_summary — advise consolidating hygiene classes repeated across assets.
Unit tests cover calibration + depth audit.
5 new doctrine meta-agents (scripts/build_methodology_v352.py → agents_md/meta/):
exploit_depth_doctrine, finding_chainer, artifact_decoder, token_auditor,
report_calibrator (meta 17→22, total 343→348).
Version bumped 3.5.1 → 3.5.2 across crates/app/installers/docs; RELEASE/README
updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
These session/runs/history files are runtime state generated during local
testing; .neurosploit/ is already in .gitignore. Untrack them so the repo
doesn't carry test artifacts.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Token/quota exhaustion no longer silently drops agents. When every candidate
model is rate-limited / out of quota, the run PARKS (keeping all state) and
prints "⏸ token/quota exhausted … PAUSED". The user can:
- wait for renewal and /continue (retry same model), or
- /model <provider:model> (or the /model selector) then /continue to switch.
Implemented via ModelPool: is_exhaustion() detection, park_exhausted() that
awaits a resume Notify, and a fallback-model slot tried first on retry. /model
queues the chosen models into a paused run's fallback so a plain /continue
resumes on them.
Findings now survive a crash/quit: each finding is checkpointed live to
.neurosploit/active_run.json; on next launch an interrupted run is recovered
into /runs (a raw report is materialized) so /results, /finding and /report
keep working.
/stop now actually halts immediately on raw/discard: one() races the in-flight
model call against the hard-cancel flag, so the CLI child (kill_on_drop) is
terminated at once instead of finishing its whole command sequence. The
validate path still soft-stops (lets validation run).
Docs: TUTORIAL documents the 3-way /stop, crash recovery and pause/continue;
/help lists /continue and the new behaviors.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Drop the legacy Python-stack settings (DATABASE_URL, HOST/PORT, RAG,
Kali sandbox, Discord/Telegram/Twilio, feature flags) that no longer
exist in the Rust harness. Keep only the provider API-key env vars the
model pool actually reads, plus the Ollama/LiteLLM base-URL overrides.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- New `litellm` provider (kind=api). Use `litellm:<model>` — model names pass
through to your gateway. No hardcoded key required (proxy may be open).
- Env-configurable base URL: LITELLM_BASE_URL (default http://localhost:4000/v1),
LITELLM_API_KEY. OLLAMA_BASE_URL override added too.
- TUTORIAL documents the LiteLLM env config.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CRITICAL BUG: truncate()/source-context slices cut strings by BYTE, panicking on
a multibyte char (e.g. '—'). The panic crashed agent tasks → task.await returned
JoinError → unwrap_or_default() → empty RunOutput. Result: real confirmed findings
(win.ini traversal, HTML injection) were silently lost, workdir was empty, report
missing. Now all string truncation is char-safe (models.rs, pipeline.rs, repl.rs).
Also:
- Background runs: /run now runs in the BACKGROUND via rustyline's ExternalPrinter
— the REPL keeps accepting commands while the engagement streams live. New
/status (live phase + progress bar + findings) and /stop (graceful). Findings
persist to history + report on completion (finalize_run ensures workdir is set
even on abort, fixing "no report file in ").
- Progress bar: agents-done/total with %, shown in /status.
- Severity colors in the live feed (Critical=red…Info=grey); confirmed vote = green.
- /help reformatted into clear aligned sections.
- TUTORIAL: document non-blocking runs, /status progress, /stop, colors.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- BUG: /auth (and /creds /focus /target /repo) with no argument CLEARED the value
instead of showing it — so typing /auth to view wiped your credential. Now no-arg
prints the current value; clear only with an explicit `clear`.
- /show now also displays API-key status (set/missing) for the selected models'
providers, and a hint of which commands edit config.
- REPL /run prints a clear "▶ RUNNING (prompt returns when done; use tui for live)"
banner before and "◀ back to the NeuroSploit REPL" after, so it's obvious the
REPL didn't disappear during a run.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Claude-Code-style @ menu: rustyline CompletionType::List so @path shows a
file/folder selection list (Tab), not inline cycling.
- /diff (/changed): shows new (+) / gone (-) findings between the last two runs.
- /retest [n]: loads a past run's target/repo and seeds a re-verify focus on its
findings → /run to check if they're fixed.
- Both added to Tab-complete and /help.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Chaining:
- agents_md/chains/ (12 multi-stage exploitation playbooks): SQLi→RCE→LPE,
SSRF→AWS-creds, SSRF→RCE, upload→RCE, upload→LFI→RCE→LPE, XSS→ATO, IDOR→ATO,
SSTI→RCE→cloud, default-creds→domain, deserialization→RCE, exposed-git→RCE,
subdomain-takeover→trusted-abuse. Each stage proven by a tool receipt before
advancing; reports chains_from edges.
- Loaded as a `chains` category (→ 329 agents). chain_round now injects the chain
recipes as a menu so the LLM applies proven multi-stage paths.
Persistence (no DB — structured state):
- Per-project `<cwd>/.neurosploit/` holding session.json (config), runs.json
(history), history.txt (readline). REPL resumes target/repo/auth/focus/models
on reopen; saves on /run and /quit.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Live findings feed: each candidate is surfaced (✦ possible finding [sev] title
@ endpoint) the moment an agent returns it, not only at the end.
- 🔔 notifications in the feed: evidence saved, phase complete (with severity
breakdown = automatic partial summary). Renderer styles notify/finding tags.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Partial observability is now first-class:
- belief.rs — property-graph world model; nodes (host/service/vuln/exploit/cred)
carry a probability, not a boolean. Bayesian observation updates; per-node
Shannon entropy; mean-uncertainty + recon-frontier. Black-box = diffuse priors
that sharpen with observation; white-box collapses toward deterministic (MDP).
- pomdp.rs — value_of_information(), decide() (recon vs exploit falls out of
belief entropy), and may_assert() — the mathematical anti-hallucination gate:
no exploitability claim while the belief is diffuse (high entropy) → observe first.
- grounding.rs — verification engine, hard rule "no claim without a tool receipt":
empirical grounding for black-box (raw HTTP/OOB/error markers), symbolic for
white-box (file:line into reviewed source). Ungrounded claims demoted + flagged
receipt_missing (feeds future reward shaping).
- pipeline.finish(): grounding gate before reporting + belief-uncertainty readout.
- bump 3.5.0 → 3.5.1; README documents the v3.5.1 belief/grounding architecture
and the infra/bandit/reward roadmap.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Streamed Claude events now tagged with the agent label (@name) so every
command/tool/file is attributable to the agent that ran it.
- Token/cost telemetry: parse usage from the stream-json result event; feed shows
per-call in/out/cost and a running total in the run summary.
- Ctrl-C during a run no longer hard-kills: it cancels cooperatively (no new
agents launch, in-flight bounded), then asks "generate report from partial
results? [Y/n]" — discard removes the run dir. Second Ctrl-C aborts.
- pool: cancel handle + is_cancelled; one()/complete_routed/chat_cli carry a label.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
REPL (rustyline Helper):
- Tab autocomplete for /commands and @filesystem-paths.
- @path attach: @file, @folder, @file:LINE / @file:START-END fold scope files /
stack traces into the agent context; /attach <path> and /context to manage.
- Multiline input: end a line with `\` to continue (validator-driven).
- /theme color|mono, /config (=/show); history (↑/↓) persists as before.
- Attachments are merged into the run's instruction context.
Install:
- setup.sh: `curl … | bash` — auto-installs Rust, clones to ~/.neurosploit,
builds release, links neurosploit into ~/.local/bin; idempotent; env-tunable.
README: v3.5.0, 🧠 (back to "neuro"), one-line install section, neurosploit-on-PATH usage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Finding enriched with owasp / mitre / kill-chain stage / exploitability /
business_impact / chains_from (attack-path edges).
- attack_graph module: derive OWASP Top 10 + MITRE ATT&CK technique + kill-chain
stage from CWE (heuristic, no extra model call); render a Mermaid attack-path
flowchart (findings grouped by stage, explicit + implicit edges) and an ASCII
kill chain for the REPL.
- enrich() runs in finish() for every engagement.
- HTML report gains an "Attack Path & Kill Chain" section (Mermaid via CDN, dark)
plus a stage/sev/OWASP/MITRE/exploitability table.
- REPL print_findings shows the ASCII kill-chain + severity summary after a run.
- models: add GPT-5.5, GPT-5.4, GPT-5.4-mini, GPT-5.3-codex, GPT-5.2.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Harness:
- ModelPool gains a progress channel (set_progress); chat_cli forwards it.
- New chat_claude_stream: drives Claude Code with --output-format stream-json and
parses the event stream live — assistant text, and tool_use blocks categorized
into tagged events (exec/danger command, read/edit file, net request/browser,
grep/glob tool). 900s bound; clear error surfacing.
- Wired set_progress into run / whitebox / greybox.
REPL renderer (render_line):
- Tagged events render as the conversation feed: tool/command/network as compact
CARDS (tool-runner visual), files/edits/AI text/states as iconized lines.
- Clear "what the AI is doing" states: reconning, planning, testing, validating,
chaining, report, complete — plus a ⚠ DANGEROUS marker for risky commands.
- Untagged harness lines mapped to the same state vocabulary.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Harness:
- Exploit-chaining round: after validation, chain confirmed findings into deeper
impact (SSRF→metadata, SQLi→dump→reuse, IDOR→ATO, file-read→secrets→RCE),
validate the new findings, merge. Wired into black-box and greybox.
- Latest top models surfaced: claude-opus-4-8, gpt-5.1/gpt-5.1-codex, gemini-3-pro.
REPL:
- Real line editing via rustyline: ↑/↓ command-history recall, Ctrl-A/E/K, paste;
Ctrl-C cancels the line, Ctrl-D exits. Command history persists to
data/repl_history.txt. Graceful plain-stdin fallback when not a TTY.
- /model with no arg → arrow-key multi-select (dialoguer); with arg accepts any
provider:model names.
- /key is model-aware: lists the providers your selected models need (set/missing)
and prompts for the missing keys; /key <prov> <key> still works.
- Run history persists to data/repl_runs.json and reloads across sessions
(/runs lists past + current; /results /report /status by run number).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- RunOutput exposes `workdir` so the session can locate reports.
- Session now records every run (RunRecord: id, mode, target, workdir, findings).
- New commands:
/runs list runs done this session (mode, target, severity counts)
/results [n] show findings of run n (default last), severity-sorted
/report [n] open the PDF/HTML report (open/xdg-open)
/status [n] print the run's status.json
/offline on|off pipeline self-test toggle (no model calls)
- Each /run prints "saved as run #n" with the quick commands.
- Verified offline: run → /runs → /results → /status all work.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- harness/creds::login(): performs the real HTTP login (POST/GET form), captures
a session Cookie from Set-Cookie or a Bearer token from the JSON body, with a
soft success check (no hard fail on 302). Redirects not followed so Set-Cookie
is visible.
- apply_creds is now async: direct material (jwt/header/cookie) used as-is; a
`login:` flow is EXECUTED to obtain a live session; on failure, falls back to
instructing the agents to log in themselves.
- --creds + --focus added to `run` (authenticated black-box) too.
- Verified live against a local mock: POST /login → 302 + Set-Cookie captured as
the auth header used on subsequent requests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- New GREYBOX mode: review a repo's source AND exploit the running app in one
pipeline — code-review findings become LEADS injected into live exploitation.
CLI: `neurosploit greybox <repo> --url <app> [--creds creds.yaml] [--focus ...]`
REPL: set both /repo and /target → greybox auto-selected.
- Credentials (harness/src/creds.rs, dependency-free YAML subset): jwt / header /
cookie, or an automated `login:` flow. Derives an auth header and/or a
"authenticate first via curl" directive injected into prompts so agents test
authenticated. --creds flag + /creds command + creds.example.yaml.
- RunConfig gains `repo`; run_engagement refactored to a Mode enum (Black/White/Grey).
- Verified offline: greybox loads creds, combines repo+URL, runs pipeline, writes report.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- New persistent interactive session (app/src/repl.rs), launched when run with no args:
banner, model selection, API-key config (/key) or subscription (/sub), then a live
session to set /target, /repo, /auth, and free-text /focus instructions (or just type
them) that STEER which agents run and how.
- Slash-commands: /model /providers /key /sub /target /repo /auth /focus /mcp /votes
/agents /show /run /quit (+ bare text = focus).
- RunConfig gains `instructions` and `auth`:
* instructions bias both LLM agent-selection and the heuristic (focus keywords →
injection/access-control/etc. agents get a strong boost)
* operator directives (focus + auth) injected into recon and exploit prompts so agents
test as an authenticated user and prioritise the requested vuln classes
- bump 3.4.1 → 3.5.0 (CLI, harness, reports, credits)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Task-based model ROUTER (recon/select prefer a fast model; exploit prefers primary; validate uses a different model than the finder)
- ReAct doctrine injected into exploit prompts (Thought→Action→Observation, token-efficient)
- Dedup: unique agents per run + findings deduped by CWE/endpoint/title (highest confidence kept)
- Token economy: recon blob capped for selector + per-agent context
- Configurable MCP: merge user mcp.servers.json into the pipeline's .mcp.json
- +54 white-box/code-analysis agents (NoSQLi, LDAP/XPath, JWT-none, Java/.NET/PHP/Go/Node/Python
specifics, SSTI, ReDoS, deserialization, etc.) → 303 agents total (78 code)
- Credits: Joas A Santos & Red Team Leaders (CLI banner, interactive header, HTML+Typst report)
- README: GitHub stars/forks badges, 60-second quick start, full API config steps, intuitive layout
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Keep only the Rust harness (neurosploit-rs/) + the agent library (agents_md/) it
loads at runtime, plus docs. Remove the Python engine, web GUIs, legacy stack,
docker, build scripts and scratch test files from THIS branch only (other
branches keep everything). Rust-focused README with Kali/Docker + tool-install
guidance and testphp/DVWA usage examples.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Root cause of empty results: models emit findings with confidence as a string
('High') or cvss as a number, but the Finding struct typed confidence as f64, so
serde failed the ENTIRE array on any mismatch -> 0 findings every run.
extract_findings now parses into serde_json::Value and coerces each field
(string/number/word), normalizes severity, and accepts qualitative confidence
(High->0.9 etc). Verified live: whitebox on a vulnerable sample now yields
validated findings (IDOR confirmed by vote).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The 'recon failed (claude subscription CLI failed: )' was a transient CLI failure
(rate limit / cold start) reported with a blank message and no retry.
- chat_cli: on non-zero exit, surface exit code + stdout (CLI writes the real
reason there, not stderr); treat empty output as an error
- pool.one(): retry up to 3x with backoff for transient failures (both
subscription and API paths)
- with_auth: cap concurrency to 3 on the subscription path — spawning many
parallel CLI processes itself trips provider rate limits
Verified: live subscription run recovers and completes recon → select → exploit
→ vote → artifacts.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Zero-dependency (stdlib http.server) front-end exposing only the essential
options — URL, backend, model, collaborator, RL + Playwright-MCP toggles — with
a live progress console. Calls neurosploit_agent directly; no npm/build.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Overhauled frontend with 2026 hacking HUD aesthetic (neon colors, glassmorphism)
- Added native support for NVIDIA NIM as a Tier 2 provider
- Fixed critical backend crashes in autonomous_agent.py and knowledge_processor.py
- Updated Kali sandbox build to Go 1.26 and fixed health check reliability
- Integrated Space Grotesk and JetBrains Mono fonts
- MD Agent system restructured: real HTTP exploitation, retry with exponential backoff, reduced concurrency (2 parallel, 2s stagger)
- Claude 4.6 model support (Opus/Sonnet) with corrected API version headers
- SmartRouter true failover with provider preference cascade
- WAFResult attribute error fix in autonomous_agent.py
- CVSS data sanitization for all vulnerability database saves
- AI recon JSON parsing robustness improvements
- rebuild.sh simplified from 714 to 196 lines
- Frontend: removed unused routes, simplified Auto Pentest page
- Agent grid: reduced max tests per agent (8→5), condensed recon prompts
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New feature: Full LLM Pentest mode where the AI drives the entire
penetration test cycle autonomously. The LLM plans HTTP requests,
the system executes them, and the LLM analyzes real responses to
identify vulnerabilities — like a human pentester using Burp Suite.
- New OperationMode.FULL_LLM_PENTEST + AgentMode enum
- _run_full_llm_pentest(): 30-round ReACT loop (plan→execute→analyze→adapt)
- 3 new prompt functions in ai_prompts.py (system, round, report)
- Anti-hallucination: findings without real evidence are rejected
- All findings routed through ValidationJudge pipeline
- FullIATestingPage updated: 4-phase UI (Recon→Testing→PostExploit→Report)
- No Kali sandbox required — uses system HTTP client directly
- Methodology injection from pentestcompleto_en.md (118KB)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Joas A Santos