agentleFS
zyphr-labszyphr-labs/bounty-harness/.claude/skills/campaign/SKILL.md

campaign

Skill1 starsChanged 3 months agoMITView on GitHub

This is the entry point for "here's a URL — go find everything you can." The deterministic engine is scripts/campaign_runner.py (invoked by bin/bb-hunt or bin/bb-run campaign hunt). The judgement parts — understanding the target, triage, and impact verification — are driven by the agent following the loop below. Safety ceiling defaults to intrusive but is auto-capped to active-safe unless a non-empty --scope-file is supplied — intrusive blackbox testing only runs against targets you have authorization for. Credentials, cookies, tokens, and…

9 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Launch: Run bin/bb-hunt <url> with the agreed --max-tier and --time-budget.

    line 30
  2. Very specific:

    Keep all of the above local-only.

    line 46

“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Campaign — Autonomous End-to-End Bug Bounty Run

Overview

The campaign skill turns a single target into a full, hands-off engagement. From one URL it runs context init, recon, domain modeling, technique matching, planning, then executes every applicable skill workflow in priority order (filtered by a safety ceiling), and finishes with reporting and readiness checks. The run is bounded by a wall-clock time budget, is resumable, and never aborts the whole campaign because one workflow failed.

This is the entry point for "here's a URL — go find everything you can." The deterministic engine is scripts/campaign_runner.py (invoked by bin/bb-hunt or bin/bb-run campaign hunt). The judgement parts — understanding the target, triage, and impact verification — are driven by the agent following the loop below.

Quick Reference

GoalCommand
Hunt a brand-new URL (cold start, does its own init)bin/bb-hunt <url> --time-budget 2h
Hunt with intrusive blackbox testing (requires authorization)bin/bb-hunt <url> --scope-file <auth> --max-tier intrusive --time-budget 3h
Hunt using an already-initialized contextbin/bb-run campaign hunt
Check progressbin/bb-run campaign status
Resume an interrupted runbin/bb-hunt <url> --no-init --resume <campaign-id>
Preview the full plan without executingbin/bb-hunt <url> --dry-run

Safety ceiling defaults to intrusive but is auto-capped to active-safe unless a non-empty --scope-file is supplied — intrusive blackbox testing only runs against targets you have authorization for. Credentials, cookies, tokens, and target responses are kept local-only, are never committed (the .bb/ and output/ trees are gitignored), and must be redacted/sanitized before any report leaves the machine.

Workflow Selection

  • New target, no prior recon, "find all vulns", "run for a while", a bare URL → start here with bin/bb-hunt.
  • Authorization confirmed / scope file in hand and you want active+intrusive blackbox coverage → --scope-file <auth> --max-tier intrusive.
  • Context already initialized for the target → bin/bb-run campaign hunt (or hunt-intrusive).
  • A previous campaign was interrupted → resume with its campaign id.
  • You only want recon/triage, not full execution → run the individual recon, domain-model, planner skills instead.

Agent Loop (how the agent drives a campaign)

  1. Authorization. Confirm the user is authorized to test the target. If they have a scope/authorization file, pass it with --scope-file to unlock intrusive testing; without it the run stays at active-safe.
  2. Understand the target (web search). Before or alongside launch, use web search and the vuln-intel and osint skills to learn what the target is: the company, its tech stack and frameworks, login/SSO providers, known CVEs for detected technologies, and any public bug bounty program scope and exclusions. Feed this into scope and into which skills to prioritize.
  3. Launch. Run bin/bb-hunt <url> with the agreed --max-tier and --time-budget. The engine handles init → recon → domain model → plan → priority-ordered execution → reporting on its own.
  4. Monitor. Poll bin/bb-run campaign status (or read .bb/campaigns/<id>/status.json and campaign.log). The run continues past individual workflow failures.
  5. Triage. As findings land under $OUTDIR, do not treat scanner output as a finding. For each candidate, run the matching vulnerability skill's verify workflow and the impact-verifier skill to rule out false positives and classify impact.
  6. Report. For confirmed, impact-verified findings, use the reporting skill (including platform-export) to produce platform-ready writeups with full evidence.

Available Workflows

  • hunt — full autonomous campaign using the already-initialized context (active-safe ceiling by default).
  • hunt-intrusive — full campaign at the intrusive ceiling; requires SCOPE_FILE authorization.
  • status — print the latest or named campaign status.
  • resume — resume an interrupted campaign by id, skipping completed workflows.

The engine's safety ceiling, time budget, skill subset, and rate limit are all configurable; see python3 scripts/campaign_runner.py hunt --help.

Evidence Required

  • Campaign state lives under .bb/campaigns/<campaign-id>/: campaign.log (timestamped, redacted run log), results.jsonl (per-workflow exit codes and timing), and status.json (live phase and counts).
  • Per-workflow findings and evidence land under $OUTDIR as defined by each individual skill.
  • Keep all of the above local-only; redact or sanitize any auth headers, cookies, bearer tokens, and PII before a report leaves the machine. Nothing in .bb/ or output/ is committed.
  • A finding is only reportable after the impact-verifier gate confirms impact and rules out false positives.

References

  • Engine: scripts/campaign_runner.py (hunt, status; --help for all flags)
  • CLI wrapper: bin/bb-hunt
  • Backbone skills: recon, domain-model, technique-kb, planner
  • Triage/report: impact-verifier, reporting, vuln-intel, osint
  • Source of truth for workflows: skill.yaml

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.