Sprint: Algorithm-Prefixed Storage Paths

Every content-addressed ID in Muse carries its algorithm as a prefix in the value (sha256:<hex>). The same principle must hold in the filesystem path: the algorithm must appear as a path segment, never as a prefix inside a filename. This makes storage algorithm-agnostic, future-proof, and Windows-safe (colons are not valid in filenames on Windows).

The object store already does this correctly: .muse/objects/sha256/<2-char>/<62-char> — every other storage area must follow the same pattern.

For nested resources (tags, releases) the path carries the algorithm of each ID independently:

.muse/tags/<repo-algo>/<repo-id>/<tag-algo>/<tag-id>.msgpack

This handles heterogeneous hashing across time — a repo created with sha256 can have tags created with blake3, and the path reflects both without ambiguity.

There is no backwards compatibility layer. All paths move to the new convention in a single atomic refactor.

repo_id: UUID → Genesis Fingerprint

repo_id is currently stored as a UUID in .muse/repo.json. UUIDs must not be used for any system-level entity. Every system entity must be identified by a content fingerprint computed at genesis: a one-time hash of the initialisation inputs (timestamp + domain + path), never re-hashed when content changes.

The code at init.py:481 already calls content_hash({"created_at": ..., "domain": ..., "path": ...}) — the correct shape. The stored value must also carry the algorithm prefix (sha256:<hex>), and all path segments derived from it must use the bare hex under an algo directory.

Agent Quick-Reference

Each row below is one atomic unit of work. An agent should complete one row per session, run only the tests listed for that row, and commit on the task branch for that row.

Detailed Grid

1 — Commits

2 — Snapshots

3 — Tags

4 — Releases

5 — Harmony Patterns

6 — Harmony Resolutions

7 — repo_id: UUID → Genesis Fingerprint

Shared Utility: split_id()

Every row above needs to extract (algo, hex) from a potentially-prefixed ID. This must live in one place.

Object Store (reference — already correct)

.muse/objects/sha256/<2-char>/<62-char> — no changes needed. object_store.py:object_path() is the canonical model every other utility method should follow. Read it before implementing any new path utility.

Execution Order

Rows must be executed in this order due to dependencies:

split_id()          ← prerequisite for everything
    ↓
commits (1)         ← no dependencies
snapshots (2)       ← no dependencies
harmony (5, 6)      ← no dependencies
    ↓
repo_id (7)         ← must land before tags/releases can finalize
    ↓
tags (3)            ← depends on repo_id format being settled
releases (4)        ← depends on repo_id format being settled

split_id and rows 1, 2, 5, 6 can be parallelized. Row 7 is the blocker for rows 3 and 4. Do not start rows 3 or 4 until row 7 is merged to dev.