Reuse existing content outputs by hash instead of re-inscribing

[shruggr]

Problem

Before a file is inscribed, nothing checks whether identical bytes already exist on chain. The cascade only reuses outputs within a repository's own history: it compares each file against the previous root's snapshot by path and cites the old outpoint when the bytes match (src/cascade.ts). The same bytes under a different path, in a fork, or in another repository get inscribed again.

The wallet does not keep the zero-sat content outputs either (they are unspendable data outputs, not basketed), so there is no local record to consult.

Proposal

Keep a local index keyed by content hash (the git blob sha is already computed for every file in src/tree.ts; sha256 would also work) mapping to the outpoint that carries those bytes. Populate it whenever the CLI publishes or fetches content, and consult it in the cascade before planning a new file output, citing the existing outpoint the same way the fork path does.

This is a wallet-side/CLI concern. The gib overlay only tracks commit heads, not files, so no server change is needed for the first version. A cross-user lookup through the overlay could come later.

Notes

• Same mechanism would apply to the SDK's ORDFS writer (buildOrdFsDirOutputs), possibly as an optional lookup hook, but nothing in the SDK is required for the CLI version.
• Reused outputs must still be reachable by the resolver from the new root (cites are 36-byte outpoints, so this is already how forks work).

🤖 Generated with Claude Code