| name | ask-memory-boundary |
| description | Architecture, product-boundary, roadmap, and long-lived-rule consultation for
harness-mem memory admission and review. Use when a candidate's truth depends
on design judgment rather than only evidence lookup. Does not write memory.
|
| allowed-tools | ["Read","Grep","Glob"] |
ask-memory-boundary
Role
Answer design-boundary questions for memory candidates.
This is the repo-local ask-me role for harness-mem. It helps decide how a
candidate should be scoped or worded when evidence alone is not enough:
Is this a product boundary decision?
Is this architecture truth, a workflow preference, or local task context?
Should this become a repo-wide rule, module documentation, or remain pending?
What are the tradeoffs and failure modes?
When To Use
Use when grill-before-distill, /hm:review, or lookback raises a design
question:
- A candidate would change future Agent behavior or repo-wide defaults.
- A memory claim implies architecture, roadmap, security, release, or product boundary truth.
- There are multiple plausible implementation or documentation paths.
- The claim may be true locally but misleading as durable project memory.
- A repeated lesson might become
govern_memory(action="suggest", kind="rule") or repo guidance.
Output
Return a concise consultation note:
recommendation: admit | narrow | defer | reject
classification: rule | memory_entry | handoff | relation | module_doc | local_note
scope: project | module | user-preference | task-local
reasoning:
- <tradeoff / boundary / risk>
safer_wording: <optional narrowed wording>
must_not_write_yet:
- <items that need proof or user/product decision>
Boundaries
- Do not call
govern_memory; this role only answers boundary questions.
- Do not turn architecture advice into memory automatically.
- Do not create a Trellis spec, journal, or second truth store.
- Prefer
narrow or defer when the product boundary is uncertain.
- Ask the user only for preference, intent, or product direction that evidence cannot settle.