Completed Agent Synchronization

SASE can exchange completed, commit-associated agents through each managed project's hidden agents sidecar. The machine-level clone lives under ~/.sase/projects/<project-key>/repos/agents; it is never exposed to launched agents or copied into numbered workspaces.

Privacy and configuration

An agents sidecar contains full chat transcripts—including prompts and responses—portable display metadata, and commit associations. Creating or pushing it therefore publishes agent data to everyone who can read the configured remote. Before running sase repo init, set the intrinsic agents sidecar's visibility: private when that data must remain restricted. Set disabled: true to opt out entirely. Synchronization never creates a remote and cannot bypass the explicit consent requested by repo initialization.

Every participating machine needs a configured machine_name. The machine hood is part of each durable agent name, such as athena.worker, so independently created names remain distinct. Run sase config init to select or create the machine identity; see machine_name and machine overlays for how the identity is stored and layered. A machine is authoritative for bundles in its own hood: pulled copies from the same machine are not imported over local artifacts, and a later local export replaces the transport copy when its commits or portable metadata change.

Portable bundle layout

The sidecar uses a versioned manifest and one directory per qualified name:

manifest.json
agents/
  athena.worker/
    meta.json
    commits.json
    chat.md

meta.json is an allowlisted projection. It excludes PIDs, workspace numbers, checkout paths, chat/output paths, and other machine-local absolute state. commits.json contains stable SHA, subject, and committed-time records. chat.md is the exact transcript. The manifest advertises a SHA-256 digest over canonical metadata and commits plus the exact chat bytes. Unsupported schemas, unsafe paths, identity mismatches, or digest failures stop that project's integration without partially importing the corrupt bundle.

Foreign bundles are reconstructed as normal terminal ace-run artifacts and standard sharded chat files. Their exact qualified names are claimed in the durable registry, so sase agent list, sase chats, name lookup, and ACE scans see them like local history. Provenance and digest markers distinguish imports from locally owned artifacts.

Commands and status

Run a mutating synchronization for all enabled projects:

sase agent sync
sase agent sync -p project-alias -p another-project

The transaction acquires a bounded per-repository lock, pulls with rebase, integrates foreign bundles, exports local bundles, commits only manifest.json and agents/, and pushes. A non-fast-forward rejection triggers one pull/recompute/commit/push retry. Conflicted rebases are aborted and reported; a failure in one project does not prevent the others from running.

Status checks use an atomic snapshot under ~/.sase/agents_sync/. A fresh snapshot is revalidated from local refs and terminal artifact markers without network access. A stale snapshot is recomputed; --refresh explicitly forces the selected fetches:

sase agent sync --check
sase agent sync --check --refresh
sase agent sync --check --json

Status reports ahead, behind, and unexported-agent counts together with missing clones/upstreams, disabled sidecars, configuration errors, and last fetch time.

Recovery

• A busy lock is a benign skip; retry after the active sync finishes.
• A missing sidecar reports not_created; run sase repo init interactively if you intend to publish agent data.
• A corrupt manifest or bundle is not overwritten automatically. Repair or restore the sidecar data, then rerun sync.
• A push or fetch failure leaves local agent history intact. Fix credentials/connectivity and retry.
• A rebase conflict is aborted before the command returns. Resolve unexpected repository state in the hidden clone, then rerun sase agent sync.