| name | earn |
| description | Anicca earn skill (GATE-0). The automaton loop calls run.sh each wake to discover and execute an earn (x402 / 0xwork / litcoin / nookplot), then VERIFIES it on-chain (tx receipt 0x1 + USDC before/after delta) and appends one line to state/earn-ledger.jsonl. One profitable wake (net>0 AND status 0x1) is the real launch gate. Use when wiring earn into the agent loop, recording an earn outcome, or verifying a profitable wake. |
earn — Anicca's GATE-0 money-maker
Spec: docs/superpowers/specs/anicca/patches/A-earn-gate0.patch.md (26 A3 / 27 A-earn).
Canonical runtime home: $LIFE_MANAGER_REPO/skills/earn/ (this dir is the committed source of truth,
shipped via the anicca-products repo so the contract lands on main/aniccaai.com; the runtime
syncs it into the agent body). The registry slot earn flips declared -> live once a real
profitable wake is verified on-chain.
What it does (no human, no Claude in the loop)
- The automaton loop invokes
run.sh each wake.
- discover mode (default): records a narrate line (no tx) so the ledger shows the wake. Never GATE-0.
- execute mode: the agent did an on-chain earn this wake →
run.sh verifies the receipt
(../_shared/lib/verify-tx.mjs → eth_getTransactionReceipt → 0x1), records the line with the derived
net_usdc, and only declares GATE-0 MET when net>0 AND status is 0x1.
Entrypoint
EARN_SOURCE=x402 ./run.sh
EARN_MODE=execute EARN_TX=0x<64hex> EARN_SOURCE=x402 \
EARN_AMOUNT=<gross_usdc> EARN_COST=<cost_usdc> EARN_TASK=<id> ./run.sh
Env (mirrors the report skill): /opt/anicca.env sourced if present; wallet privkey var named
by PKVAR (default BLOCKRUN_WALLET_KEY) derives the wallet address for the USDC delta proof.
Optional: BASE_RPC_URL, USDC_ADDRESS, EARN_LEDGER, WAKE_ID.
Ledger (state/earn-ledger.jsonl, append-only)
{"ts":...,"wallet":"0x..","source":"x402","task":"...","earn_usdc":0.5,"cost_usdc":0.05,
"net_usdc":0.45,"tx":"0x..","status":"0x1","wake":"..."}
../_shared/lib/ledger.mjs never rewrites prior lines (immutability). isProfitable() is the single
GATE-0 classifier. narrate-only lines (no tx) never count.
P1 fail-closed CUMULATIVE guard (spec §3 P1 / §4) — the one-line pattern for every earn skill
isProfitable() only judges ONE pass. ../_shared/lib/earn-guard.mjs adds the CUMULATIVE layer the
spec's hard invariant actually needs: 稼ぐ額 >>> 使う額 over TIME, not per pass. It sums net_usdc
across the whole ledger for a {wallet, source} scope (per-skill) AND a {wallet}-only scope
(per-agent, every source combined) and HALTs — fail-closed — the instant either cumulative total
would go negative, or the instant any matching line's earn_usdc/cost_usdc/net_usdc isn't a real
number (a malformed line is treated as unverifiable, never as harmless/zero).
record.mjs's JS API (record()) already returns this as { line, profitable, halt } — nothing else
to wire if a skill only ever calls record() directly. For a shell entrypoint, add ONE line mirroring
the existing kill-switch idiom (polymarket-trade/run.sh's if [ -f "$SKILL_DIR/KILL" ]), checked
BEFORE doing anything that wake:
if ! node "$HERE/../_shared/lib/earn-guard.mjs" check "$WALLET" "$SOURCE" "$LEDGER"; then
echo "P1 GUARD: cumulative net breach — HALT (fail-closed), skipping wake."; exit 0
fi
$SOURCE may be "" to check the per-agent (wallet-wide) scope only, when the wake doesn't yet know
which specific source it will use this pass. Exit 0 = solvent, exit 1 = HALT (reason on stderr).
Do NOT gate this call on $WALLET being non-empty (e.g. if [ -n "$WALLET" ] && ! node ...).
That was FIND-A (adversary round 2): when identity resolution fails, $WALLET is empty, the &&
short-circuits, and the guard never runs at all — the wake proceeds and its loss gets recorded under
a literal "unknown" wallet, invisible to every real-wallet-scoped check forever. ALWAYS call the
guard unconditionally; earn-guard.mjs check itself fail-closed HALTs (exit 1, reason
missing-wallet) on an empty/missing wallet — a broken identity is exactly when the wake must NOT
proceed, never a reason to skip the check.
Wired today:
earn/run.sh — the guard clause above, checked once at the very top (wallet-wide scope; every
strategy branch — 0xwork/yield/swap/hl/token — inherits it for free).polymarket-trade/redeem.py — after EVERY redeemed condition, check_cumulative_halt() calls the
same CLI (scope {wallet: DEPOSIT_WALLET, source: "polymarket-redeem"}); on HALT it writes
polymarket-trade/KILL, so the trading entrypoint's UNCHANGED existing kill-switch check stops the
NEXT pass — zero changes needed to polymarket-trade/run.sh itself.
Not yet wired (these don't append to state/earn-ledger.jsonl yet, so there's nothing cumulative
to guard): hl-trade, sol-trade, x402-sell. Once any of them starts calling record.mjs/
record_ledger_line-style writes, add the one-liner above at its own pass boundary.
UBI (spec 28 §0) — share OWN earnings with AI + human recipients
After a wake records PROFITABLE (real external USDC, isProfitable), run.sh calls
../ubi/distribute-ubi.mjs: it sends UBI_SHARE_BPS (default 10.00%) of THIS wake's net, split equally,
to sibling-AI child wallets (self/spawn/state/children.jsonl) + a human allow-list
(UBI_HUMAN_WALLETS / state/ubi-recipients.json), via one ERC20 USDC transfer each
(0xa9059cbb, USDC 0x8335…2913, 6dp — ctx7-verified). It funds ONLY from Anicca's OWN wallet and
touches ZERO user identity (earn-side of the spec 28 §3 wall). Idempotent per wake; never sends
more than earned; UBI_DRY_RUN=1/below-min/no-recipients → record dry/skipped (NO fake send).
Audit trail: append-only state/ubi-ledger.jsonl ({kind:"ubi",wake,outcome,txs:[{to,tx,status}]})
— separate from the earn ledger so the GATE-0 classifier is untouched.
Verify (independent agent)
node -e "import('../_shared/lib/verify-tx.mjs').then(m=>m.receiptStatus('0x<tx>')).then(console.log)"
node -e "import('../_shared/lib/usdc.mjs').then(m=>m.usdcBalance('0x<wallet>')).then(console.log)"
node --test lib/__tests__/*.test.mjs __tests__/*.test.js
node --test ../_shared/lib/__tests__/earn-guard.test.js
Acceptance: wallet USDC after-before>0 for the wake + a ledger line whose Base receipt is 0x1.
Narration alone FAILS (HARD 0.24/0.31).
