Repository navigation
Conversation
Refs bytefolk#356. Design only. No runtime. Gate D0 still required before implementation.
PeterGuy326
left a comment
There was a problem hiding this comment.
Reviewed head-to-tail. Design-only doc, no runtime. Branch heads are per-hire, cross-hire head writes are rejected, merges create new segments plus a merge record instead of clobbering, and the handoff package carries only refs the successor can already read (no leftover grants, no foreign private memoryIds). Digests are explicitly illustrative. Consistent with #364 and #328 re-evaluation-at-recall. One cosmetic note: the AC sections appear in order AC-001, AC-003, AC-002 in the file; harmless.
|
@sun-970 Summary for your six memory-plane PRs (#364, #369, #370, #371, #372, #373): all were reviewed, approved, and merged to main via in-repo recreations #375, #377, #378, #379, #380, #381 — file blobs byte-identical to your fork heads, commits authored as you, all required checks green before each merge. The originals were closed because fork pull requests cannot satisfy the required CodeQL check name in branch protection on this repository (structural: every merge in this repo's history used an in-repo branch). Nothing needs fixing on your side — your design docs are all on main. Going forward, please push branches directly to bytefolk/roleweave and open PRs from there (same as you did for #342/#343) so they can go through the normal protection checks. |
Tracking record
Refs #356
Summary
Design-only RW-3: per-agent branch heads, required merge records, handoff packages carry checkpoint + receipt refs (not leftover grants or transcripts). Agent B cannot silently overwrite Agent A’s checkpoint. Successor permissions re-evaluated at recall (#328).
No runtime. Blocked on Gate D0 and DE-2.
Acceptance criteria
docs/design/memory-branch-handoff-v1.mdValidation ledger