One memory engine addressed by four typed scopes (global, agent, user,
agent-user), CRUD with retained revisions, one write path, and no identity
resolution inside the engine. Authority:
docs/prds/memory-engine-unification.md (Approved 2026-09-16, supersedes the
draft memory-turn-context.md).
Problem: two live memory stores for one assistant, split by writer, neither
reaching a prompt. internal/memory (Manager + builtin.Provider, MEMORY.md/
USER.md) is written by the chat agent's memory_edit tool and never read back.
internal/domain/memory + internal/infrastructure/memory is written by the
session curator under an Identity that is really a session id. The contract has
no Update and no scope model. SenderID is never persisted, so half the channels
carry no user identity at all.
Outcome: internal/domain/memory is the only engine; Scope{Kind,Agent,User} is
the addressing scheme; the chat turn reads its subject's scopes synchronously
before the prompt and writes through an id-addressed tool; the curator writes
agent-user scope; the scanner moves into the engine; internal/memory is deleted.
Not determined (left open inside the PRD): group-chat extraction, concurrent
scope-file writers, unbounded HISTORY.md, size bounds, and whether the dashboard
should carry a user identity.
Findings: memory-4 and its dependants (.local/issue-tracker), for which the
split above is the root cause.
Design
docs/prds/memory-engine-unification.md
Acceptance criteria
The PRD's nine acceptance criteria hold, all asserted on behaviour or planted content (docs/prds/memory-engine-unification.md, 'Acceptance criteria').
Close reason
All 5 child slices (2ent.1 through 2ent.5) are complete and merged to main; internal/memory is deleted.
Imported from beads archie-core-2ent · GitHub #853 · type epic · priority P1 · created 2026-09-17 · closed 2026-09-19
One memory engine addressed by four typed scopes (global, agent, user,
agent-user), CRUD with retained revisions, one write path, and no identity
resolution inside the engine. Authority:
docs/prds/memory-engine-unification.md (Approved 2026-09-16, supersedes the
draft memory-turn-context.md).
Problem: two live memory stores for one assistant, split by writer, neither
reaching a prompt. internal/memory (Manager + builtin.Provider, MEMORY.md/
USER.md) is written by the chat agent's memory_edit tool and never read back.
internal/domain/memory + internal/infrastructure/memory is written by the
session curator under an Identity that is really a session id. The contract has
no Update and no scope model. SenderID is never persisted, so half the channels
carry no user identity at all.
Outcome: internal/domain/memory is the only engine; Scope{Kind,Agent,User} is
the addressing scheme; the chat turn reads its subject's scopes synchronously
before the prompt and writes through an id-addressed tool; the curator writes
agent-user scope; the scanner moves into the engine; internal/memory is deleted.
Not determined (left open inside the PRD): group-chat extraction, concurrent
scope-file writers, unbounded HISTORY.md, size bounds, and whether the dashboard
should carry a user identity.
Findings: memory-4 and its dependants (.local/issue-tracker), for which the
split above is the root cause.
### Design
docs/prds/memory-engine-unification.md
### Acceptance criteria
The PRD's nine acceptance criteria hold, all asserted on behaviour or planted content (docs/prds/memory-engine-unification.md, 'Acceptance criteria').
### Close reason
All 5 child slices (2ent.1 through 2ent.5) are complete and merged to main; internal/memory is deleted.
---
_Imported from beads `archie-core-2ent` · GitHub #853 · type epic · priority P1 · created 2026-09-17 · closed 2026-09-19_
sam
added the highepic labels 2026-10-27 13:38:41 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
One memory engine addressed by four typed scopes (global, agent, user,
agent-user), CRUD with retained revisions, one write path, and no identity
resolution inside the engine. Authority:
docs/prds/memory-engine-unification.md (Approved 2026-09-16, supersedes the
draft memory-turn-context.md).
Problem: two live memory stores for one assistant, split by writer, neither
reaching a prompt. internal/memory (Manager + builtin.Provider, MEMORY.md/
USER.md) is written by the chat agent's memory_edit tool and never read back.
internal/domain/memory + internal/infrastructure/memory is written by the
session curator under an Identity that is really a session id. The contract has
no Update and no scope model. SenderID is never persisted, so half the channels
carry no user identity at all.
Outcome: internal/domain/memory is the only engine; Scope{Kind,Agent,User} is
the addressing scheme; the chat turn reads its subject's scopes synchronously
before the prompt and writes through an id-addressed tool; the curator writes
agent-user scope; the scanner moves into the engine; internal/memory is deleted.
Not determined (left open inside the PRD): group-chat extraction, concurrent
scope-file writers, unbounded HISTORY.md, size bounds, and whether the dashboard
should carry a user identity.
Findings: memory-4 and its dependants (.local/issue-tracker), for which the
split above is the root cause.
Design
docs/prds/memory-engine-unification.md
Acceptance criteria
The PRD's nine acceptance criteria hold, all asserted on behaviour or planted content (docs/prds/memory-engine-unification.md, 'Acceptance criteria').
Close reason
All 5 child slices (2ent.1 through 2ent.5) are complete and merged to main; internal/memory is deleted.
Imported from beads
archie-core-2ent· GitHub #853 · type epic · priority P1 · created 2026-09-17 · closed 2026-09-19