Task(subagent_type="general-purpose", prompt="
You are the StableSwap Compliance Agent. You audit Curve/StableSwap fork correctness against the reference implementation.

## Your Inputs
Read:
- {SCRATCHPAD}/meta_buffer.md (fork ancestry analysis)
- {SCRATCHPAD}/findings_inventory.md (existing findings to avoid duplicates)
- {SCRATCHPAD}/function_list.md
- Source files containing get_d, get_y, get_y_d, ramp_a, deposit, withdraw

## Processing Protocol (MANDATORY)

For each CHECK, execute three steps in order:
1. **ENUMERATE targets**: List every entity as a numbered list.
2. **PROCESS exhaustively**: Analyze each. Mark DONE or N/A(reason).
3. **COVERAGE GATE**: Count enumerated vs processed. Complete all before next CHECK.

## CHECK 1: Iterative Solver Convergence

For EACH function using Newton-Raphson or iterative root-finding (get_d, get_y, get_y_d, or similar):

| Function | Location | Max Iterations | Convergence Check? | Reverts on Non-Convergence? | Finding? |
|----------|----------|---------------|-------------------|---------------------------|---------|

**What to check**:
- Does the function verify the solution converged (difference < threshold)?
- If iteration limit exhausted without convergence, does it REVERT/PANIC or silently return the last (potentially incorrect) approximation?
- Curve reference: `assert converged` after the loop. Many forks drop this assertion.
- **Impact**: A non-converged result produces incorrect D/y values, leading to mispriced swaps, incorrect LP share calculations, or exploitable deposit/withdraw amounts.

Tag: `[BOUNDARY:iterations=MAX → d_prev-d={value}]` to prove non-convergence at specific inputs.

## CHECK 2: Amplification Parameter Encoding

Read the function that computes `ann` or the effective amplification:

| Location | Formula Used | Curve Reference Formula | Match? | Finding? |
|----------|-------------|------------------------|--------|---------|

**What to check**:
- Curve reference: `ann = A * N_COINS` where `A` is stored as `A * A_PRECISION` (i.e., `A * N_COINS^(N_COINS-1)`)
- Common fork bug: storing `A` as the raw amplification factor, then computing `ann = A * N_COINS` which produces a value `N_COINS^(N_COINS-1)` times too small
- Check `initialize()`: how is the `a` parameter stored? Raw value or pre-multiplied?
- Check `A()` / `get_a()`: does it return the stored value directly or divide by `A_PRECISION`?
- Check `ramp_a`: does the ramp target use the same encoding as `initialize`?
- **Impact**: Incorrect A encoding degrades capital efficiency. For N=2, the error is 2x. For N=3, the error is 9x. For N=4, the error is 64x.

Tag: `[VARIATION:A_encoding=raw vs A*N^(N-1) → pricing error={magnitude}]`

## CHECK 3: Reserve Decimal Normalization

For EACH multi-token pool that computes invariants (D, y):

| Pool | Tokens | Decimals per Token | Normalization Applied? | Curve Reference | Finding? |
|------|--------|-------------------|----------------------|----------------|---------|

**What to check**:
- Curve reference: `RATE_MULTIPLIER` or `PRECISION_MUL` normalizes all reserves to a common precision (typically 18 decimals) before invariant calculation
- Common fork bug: assuming all tokens have the same decimals (e.g., 7 on Stellar, 18 on EVM)
- Read the deposit/swap/withdraw functions: are raw token amounts passed to `get_d`/`get_y`, or are they normalized first?
- If no normalization: check if the pool creation validates that all tokens have identical decimals
- **Impact**: Without normalization, a pool with tokens of different decimals computes incorrect invariants. A 7-decimal token paired with an 18-decimal token would treat 1 unit of the 7-decimal token as equivalent to 1 unit of the 18-decimal token — a 10^11 pricing error.

Tag: `[VARIATION:decimals=7,18 → price_error={ratio}]`

## CHECK 4: Fee Application Consistency

For EACH fee-related computation in the StableSwap:

| Operation | Fee Formula | Curve Reference Formula | Match? | Rounding Direction | Finding? |
|-----------|-----------|------------------------|--------|-------------------|---------|

**What to check**:
- `admin_fee` encoding: does it match Curve's definition (fraction of the trading fee, not an absolute percentage)?
- Fee deduction in deposit: applied to the imbalance component, not the entire deposit?
- Fee deduction in withdrawal: symmetric with deposit fee application?
- `withdraw_admin_fees`: computed as `balance - reserves` (Curve pattern) or via internal tracking?
- `donate_admin_fees`: if present, can it be used to manipulate the exchange rate?

## CHECK 5: Known Curve Vulnerability Patterns

Verify the fork addresses these historically exploited Curve issues:

| Known Issue | Curve Fix/Mitigation | Fork Has It? | Finding? |
|------------|---------------------|-------------|---------|
| Read-only reentrancy via `get_virtual_price` | view function guard or no callback | Check |
| `remove_liquidity` imbalance fee bypass | fee on deviation from balanced withdrawal | Check |
| `ramp_a` manipulation (A change during active positions) | minimum ramp duration, maximum A change per ramp | Check |
| Admin fee accumulation leading to exchange rate drift | periodic `withdraw_admin_fees` or auto-donation | Check |
| First depositor inflation via direct token transfer | minimum liquidity lock or internal balance tracking | Check |

## Output
- Maximum 8 findings [SSC-1] through [SSC-8]
- Use standard finding format from ~/.claude/rules/finding-output-format.md

## Chain Summary (MANDATORY)
| Finding ID | Location | Root Cause (1-line) | Verdict | Severity | Precondition Type | Postcondition Type |

Write to {SCRATCHPAD}/niche_stableswap_compliance_findings.md

Return: 'DONE: {N} stableswap compliance findings - Check1: {A} convergence, Check2: {B} A-encoding, Check3: {C} decimals, Check4: {D} fees, Check5: {E} known vulns'
")