Done. The repository now holds merkledrop, a dependency-free Foundry command-line tool, committed as 1a5c5d1 by surfer77 with a clean message. The working tree is clean apart from the untracked test/scratch/ directory, which the harness deletes.

What the tool does. It reads an airdrop allocation list (<address>,<amount> lines) from standard input and produces the Merkle root, a single account's claim proof, or a full JSON manifest of every claim. The tree is byte-for-byte compatible with OpenZeppelin's StandardMerkleTree for (address, uint256), verified against that package in node for the five-entry fixture and for a 3,900-entry list.

Layout. Five pure libraries under src/ (string conversions, stable merge sort, allocation parsing, the Merkle tree, and command dispatch with JSON rendering), a thin forge script entry point, and a POSIX shell launcher in bin/merkledrop that passes stdin and arguments through. The README covers input rules, commands, output shapes, build and run steps, and declares the check command. A Makefile exposes it as make check.

Verification.

Check	Result
forge build && forge test (52 tests, 6 suites)	pass, no warnings
Same suite under solc 0.8.13 and 0.8.20	pass
Clean git clone of the commit, build and test	pass
Launcher on 3,900 allocations	root in about 1 s, manifest in about 4 s

The tests cover each module in isolation, the entry point end to end, and the unhappy paths at least as thoroughly as the happy ones: empty input, missing or extra fields, malformed addresses, bad EIP-55 checksums, non-decimal, zero and overflowing amounts, duplicate accounts, oversized input at and just past the 256 KiB limit, unknown commands, and wrong arguments. Two fuzz tests hold invariants across random inputs: sorting yields a stable sorted permutation, and every proof verifies while altered leaves and reused proofs fail.

Two design decisions worth knowing. Forge's readFile cannot read a piped stdin (it locks the file), so the launcher reads stdin itself and hands it to the script as an argument, which is what makes the entry point directly testable. Foundry 1.x ships with the optimizer off and a 1 B gas cap, so foundry.toml enables the optimizer and lifts the gas cap; the input size limit and forge's memory limit bound a run instead.