Release v4.1.0 — Slim Orchestrator, On-Demand Loading · Gabson0x/bountyforge

v4.1.0 — Slim Orchestrator, On-Demand Loading (2026-09-16)

Context-budget release. SKILL.md 2,992 → 872 lines (162KB → 65KB) without content loss.

SKILL.md slimmed 3.4×: doctrine, mode selection, skill chain, orchestration, and hot-path sections stay inline; deep checklists moved verbatim to 9 new references/ files and load on demand.
Banner + AUTO-UPDATE SYSTEM retained — v4.0.0 session-start behavior unchanged.
Frontmatter description trimmed (~230 words → 2 sentences). Trigger coverage retained; always-resident prompt cost reduced.
Removed duplicate skills/pashov/fizz (byte-identical to skills/fizz).
bountyforge.skill regenerated.

CONTEXT BUDGET — explicit load-on-demand rules; references and sub-skills load only when their trigger fires.
REFERENCE MAP — topic/trigger → exact file to load.
EXTERNAL SKILL PACKS — cross-reference to the optional pentest-skills pack (crazyMarky, Apache-2.0): recon-subdomain, recon-port-scan, recon-fingerprint, recon-dir-scan, exploit-sqli, exploit-xss, exploit-lfi, exploit-file-download, pentest-report, results-storage. Referenced by skill name (public-safe); BountyForge gates/report format take precedence.
9 new reference files extracted verbatim:
- references/web3-quickstart.md — multi-chain + grep arsenal
- references/mode-modules.md — godmod / fuzz / x-ray / meme / storage
- references/ab-chains.md — H100 proven A→B chains
- references/recon-and-leaks.md — Phase 1 recon + credential/source leaks
- references/learn.md — Phase 2 disclosed-report pipeline + threat model
- references/web-hunting-playbooks.md — all bug-class checklists
- references/validation.md — SC reasoning, 7-Question Gate, triage, CVSS
- references/reporting.md — report format, formulas, reject lists
- references/python-tooling.md — tools/ command reference