feat(local-env): end-to-end stress harness covering all merged PRs (132-138)

[tangletools]

Summary

Single-file local stress harness at scripts/local-env/stress-test.sh plus a runbook at scripts/local-env/STRESS-TEST.md. Boots a fresh anvil + LocalTestnet.s.sol deployment in subscription mode and walks 17 ordered economic checks against the merged-PR surface. Average wall time on a warm-cache checkout: ~40-85s.

• Idempotent: re-runs tear down anvil, broadcast artifacts, indexer docker volumes, and persisted state.
• Per-step [OK] / [FAIL] lines with timing + a one-line headline metric (e.g. draw1=80909090909090909 wei, actualSlashed=3150000000000000000 wei).
• Single final RESULT: N / 17 OK in Xs summary; exit 0 ⇔ all green.
• Optional --with-indexer / --with-dapp / --with-operator for the off-chain side processes; none required for the 17 protocol-surface assertions.

Steps + which PR each step validates

#	Step	Validates
01	prerequisites	tools + soldeer deps present
02	idempotent cleanup	re-runnable
03	anvil + contracts deployed	LocalTestnet.s.sol runs in subscription mode (#132 baseline)
04	resolve deployed addresses	broadcast artifact parses to Tangle + staking proxies
05	optional services launched	indexer / dApp / operator flags are best-effort
06	operator2 registered	isOperatorRegistered(0, op2) (#132 setup)
07	subscription service Active	isServiceActive(0) + baseline pinned (#132)
08	operator on service operator set	isServiceOperator(0, op2) — ApprovalsViews split clean (#133)
09	escrow funded > 0	getServiceEscrow(0).balance > 0
10	billSubscription #1	multi-asset TWAP draw executes (#133)
11	second staker grows pool	MultiAssetDelegation.depositAndDelegate lifts getOperatorDelegatedStake (#134 O(1) aggregate)
12	billSubscription #2 (post-stake)	bill still accurate after pool growth (#133 + #134)
13	proposeSlash + executeSlash	7-day dispute window + TIMESTAMP_BUFFER respected, actualSlashed > 0 (#134 share-pool slashing)
14	operator stake reduced post-slash	getOperatorDelegatedStake strictly drops (#134)
15	claimRewardsAll (native)	native sweep zeroes pendingRewards(op2, 0x0)
16	griefing token skipped	RevertingTransferERC20 seeded via anvil_setStorageAt into _pendingRewards[op2][grief] + _pendingRewardTokens[op2]; claimRewardsAll iterates, isolates the revert, emits RewardsClaimSkipped and leaves the entry intact for retry (#136)
17	indexer entities present	with --with-indexer: at least one row each in Operator, Service, ServiceOperator, SubscriptionBilling, PaymentDistribution, RewardClaim, RewardsClaimSkip (#138), SlashProposal, OperatorPoolSlash (#138)

Implementation notes worth flagging at review:

• vm.store from a broadcast forge script does not propagate to anvil. The griefing seed (step 16) needs anvil_setStorageAt to actually mutate live storage — so the slot computation runs from bash and the forge helper (script/StressGriefingSeed.s.sol) only deploys the reverting ERC20.
• Storage slots are pinned from forge inspect Tangle storage-layout at this commit (_pendingRewards at slot 44, _pendingRewardTokens at slot 64). If those move, step 16 will print the discovered value so the slot constant can be re-derived. Runbook documents the procedure.

Test plan

• Cold-cache full run completes 17/17 (~85s).
• Warm-cache subsequent runs complete 17/17 (~40-50s).
• Three consecutive idempotent runs all 17/17 (no state bleed).
• --with-indexer flag runs 17/17 including indexer entity checks. (Out of scope for this PR — the worktree's pnpm/envio bootstrap is finicky; runbook documents the workaround.)
• --with-dapp / --with-operator smoke (these are best-effort; failures do not affect the harness pass/fail).

[tangletools]

🔍 Reviewing f7695d20

Pass	Status	ETA
Kimi Code K2.6	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #143 · model: kimi-for-coding · started 2026-05-16T10:41:32Z}