01 / OPERATING MODELOPEN SOURCE / AUTHORIZED TESTING

BOUNTYHARNESS / EXECUTION + EVIDENCE

Execution and
evidence for
security research.

BountyHarness is an execution and evidence system for authorized AI-assisted security research. It initializes scope, dispatches registered workflows, enforces safety tiers, preserves run traces, verifies impact, and leaves a reportable evidence set on disk.

MIT licensed / local-only evidence / macOS + Linux

FIG. 01CAMPAIGN CONTROL FLOWLOCAL / TRACEABLE
target + scope -> controlled execution -> evidencev3.1 / thin harness / fat skills
INSTALL / UNIX SHELLmacOS / Linux
curl bootstrap
$curl -fsSL https://raw.githubusercontent.com/Mr-Neutr0n/bounty-harness/main/install.sh | sh -s --
46skill packages
99registered tools
4safety tiers
MITopen source license
02 / CONTROL SURFACE

Every transition has a file, a gate, or a trace.

The shell stays small. Context initialization, workflow dispatch, registry checks, safety enforcement, and trace capture are explicit. The security logic lives in the skill packages.

01

Context and scope

Initialize a target, program, scope file, output directory, rate limit, and user agent before a workflow runs.

bb-init / bb-validate
02

Workflow selection

Run a named workflow from a skill's `skill.yaml`, or let the campaign runner move from recon to plan to execution.

bb-run / bb-hunt
03

Governance and trace

Safety tiers, scope gates, rate limits, and the circuit breaker constrain execution. Every run records timing, tier, tools, and exit code.

bb-tools / .bb/traces
04

Impact and evidence

Candidate output is not a finding. Verification and reporting assemble reproducible artifacts under the current output directory.

impact-verifier / reporting
03 / INSPECTABLE RUN

The useful unit is the run, not the prompt.

Commands remain ordinary shell commands. The surrounding context, workflow definition, trace, and evidence paths make the work reviewable by another researcher.

EXAMPLE / DOCUMENTED COMMANDSNO TARGET REQUESTS HERE
$ bb-init example.com --program example --scope-file scope.txt
  writes .bb/context.env
  writes .bb/context.json

$ bb-run recon passive-subdomains
$ bb-run domain-model profile
$ bb-run planner generate-plan-safe

$ bb-run reporting batch-generate
  reads candidate results and evidence
  writes report and evidence manifest
EVIDENCE / REQUIRED MATERIALLOCAL OUTPUT
REQUEST
evidence/<finding>/request.txt
RESPONSE
evidence/<finding>/response.txt
PROOF
evidence/<finding>/poc.sh
VISUAL
evidence/<finding>/screenshot.png
MANIFEST
timestamp and tool versions

The repository treats scanner output as a candidate. Impact verification and false-positive review happen before reporting.

04 / SAFETY MODEL

Safety is part of the execution path.

Workflows declare risk. The harness checks scope and applies the highest required tier instead of treating every command as equivalent.

passiveread-only discovery or local analysisgate: none
active-safetarget interaction that should remain read-onlygate: rate limit
intrusivepayloads, fuzzing, WAF-triggering scansgate: non-empty scope file
destructive-manualdata modification or privilege changesgate: human approval

Scope gate. Without a non-empty scope file, an intrusive campaign is automatically capped to `active-safe`.

Circuit breaker. Consecutive workflow failures put a target into cooldown until the breaker is reset.

Trace. Run records stay under `.bb/traces/`; target data and evidence paths are local and gitignored.

05 / SKILL CATALOG

Skills are inspectable packages.

Each package carries its router, executable workflow registry, scripts, runbooks, payloads, evidence rules, and standards context. The harness dispatches; the package contains the method.

Browse the 46 packages
RECON + PLANNING

recon / domain-model / technique-kb / planner / coverage / asset-graph / traffic-corpus / campaign

WEB VULNERABILITIES

xss / sqli / ssrf / rce / file-upload / cors-csrf / race-condition / http-protocol / nuclei-scanner

IDENTITY + WORKFLOW

auth / api / persona / cross-account / business-logic / impact-verifier / scope-manager

SPECIALIZED AUDITS

cicd-security / graphql-audit / web3-audit / meme-coin-audit / identity-domain / binary-analysis

RESEARCH + REPORTING

osint / vuln-intel / standard-catalog / reporting / program-memory / triage-validation / enrichment

END / START LOCAL

Inspect the workflow. Run what fits the scope.

MIT licensed. macOS and Linux. No account and no telemetry. Use BountyHarness only on systems you own or are explicitly authorized to test.