Summary: Over the past year I built and open-sourced a stack of AI agent safety infrastructure, all verified like flight software. This proposal asks for retroactive funding: the work is done, the code is public, the tests are green. A lump sum recognizes the delivered public good and sustains its maintenance.

What has been achieved: The stack is already built and open: agent-cage, a fail-closed safety library for AI agents (guard() returns allow, ask, or deny; frozen policies; separation-of-duties ticket approvals; irreversible kill switch; thread-safe budget caps; hash-chained, HMAC-signed audit log; 133/133 tests green in clean-room venvs; an adversarial audit found 15 issues (4 critical), all remediated and re-verified; github.com/prayingperceptions/agent-cage). The Jubilee Engine verification gauntlet: a 24/7 build loop (plan, vet, execute in isolated worktrees, verify, open PR); every proposed change must pass typecheck, tests, lint, dependency audit, Slither/Aderyn security scans, and an adversarial red-team before a human sees it; the engine is PR-only by design, it never deploys and never moves funds, and money-adjacent work is hard-gated to a human (github.com/Jubilee-Protocol/jubilee-agent). Proof it works on real money: a keyless bridge that let an autonomous agent earn its first real Bitcoin, watching USDC payments on Base and converting to sats at the live BTC price, with no exchange keys and no brokerage account. Supporting tooling, all MIT-licensed: swarm-kit (three agent tools, 109/109 tests, a 300-task fanout that absorbed 51 crashes and passed 299; github.com/prayingperceptions/swarm-kit), rugshield (on-chain-verified token-risk scanner with first real paid scans), and a token-risk API with verified per-scan on-chain settlement. The goals now: keep the repositories maintained and hardened, complete a professional third-party audit of agent-cage, and keep the verification loop running, through nights-and-weekends stewardship alongside full-time work.

How to know it will work: The third-party audit is published with its findings remediated, the repositories stay maintained with green builds, and the stack remains available as open reference infrastructure for fail-closed agent design.

Funding use: Continued maintenance of the repositories, a professional third-party security audit of agent-cage, and compute and API costs for the 24/7 verification loop. Every output stays open source.

Team: Solo builder. The delivered stack described above is the track record: everything shipped, verified in clean-room environments, with adversarial audits completed and remediated.

Variables, causes & outcomes: The core work is already delivered and open, so the failure mode is limited: without funding, the professional audit doesn't happen and maintenance slows. The public goods remain available regardless.

Capital raised: Zilch. I saw a gap and built based on a series of needs. Sweat equity. Every bit helps. Thank you!