Summary
Mneme should steal six6's skeleton, not its unsafe runtime behavior.
The useful part is the repo/product shape:
- clear layering
- canonical artifacts
- one obvious operator CLI
- install/migration docs
- cron/systemd examples
- tiny sample workspace
This would make Mneme less smart-but-fuzzy and more deployable.
Why
Mneme already has strong memory logic, but the operator story is still scattered.
A clearer package shape would help with:
- onboarding
- repeatable deployment
- docs clarity
- automation entrypoints
- making Mneme feel like a product instead of a bag of scripts
What to steal
1. Clear layering
Adopt a stronger split like:
protocol/ — schemas, canonical artifact definitions, contractsruntime/ — operator entrypoints / orchestration wrappersdistribution/ — install, migration, cron/systemd, deployment examples
2. Canonical artifacts
Make the expected files/directories more explicit:
- raw evidence
- compiled outputs
- reviewed outputs
- runtime handoff artifacts
- health/status outputs
3. Single operator CLI
Instead of many separate scripts being the public face, add one obvious runtime entry.
Example shape:
mneme checkmneme ingestmneme compilemneme driftmneme run- later:
mneme doctor / mneme batch
4. Distribution layer
Add first-class:
- installer/bootstrap flow
- migration docs/scripts
- cron example
- systemd example
5. Tiny sample workspace
A very small sanitized sample helps people understand the expected structure immediately.
What not to steal
Do not copy six6's weak runtime model:
- unsafe whole-file rewrite loops
- silent success on failure
- weak ad hoc validation
- queue-like flows without locking/merge safety
- file-driven mutations with weak guardrails
Suggested acceptance criteria
- Mneme repo structure makes protocol/runtime/distribution boundaries obvious
- one top-level operator command becomes the default UX
- deployment examples exist for cron/systemd
- sample workspace exists and is sanitized
- validation is schema-backed, not hand-wavey
- failure propagation is strict (no fake green runs)
Nice follow-up
If this lands, the next step is to define the minimal public CLI contract and decide which current scripts stay internal vs public.
Summary
Mneme should steal six6's skeleton, not its unsafe runtime behavior.
The useful part is the repo/product shape:
This would make Mneme less smart-but-fuzzy and more deployable.
Why
Mneme already has strong memory logic, but the operator story is still scattered.
A clearer package shape would help with:
What to steal
1. Clear layering
Adopt a stronger split like:
protocol/— schemas, canonical artifact definitions, contractsruntime/— operator entrypoints / orchestration wrappersdistribution/— install, migration, cron/systemd, deployment examples2. Canonical artifacts
Make the expected files/directories more explicit:
3. Single operator CLI
Instead of many separate scripts being the public face, add one obvious runtime entry.
Example shape:
mneme checkmneme ingestmneme compilemneme driftmneme runmneme doctor/mneme batch4. Distribution layer
Add first-class:
5. Tiny sample workspace
A very small sanitized sample helps people understand the expected structure immediately.
What not to steal
Do not copy six6's weak runtime model:
Suggested acceptance criteria
Nice follow-up
If this lands, the next step is to define the minimal public CLI contract and decide which current scripts stay internal vs public.