Configure verify skill for a cheaper model

The commit and verify skills were burning the full parent-conversation
context on every routine invocation. Both now set model: haiku in
frontmatter so that context is billed at the cheap rate instead of
whatever model the main session is running.

context: fork was tried on both skills to avoid shipping the parent
context at all, but reverted: a bare "/commit" or "commit this" with no
explicit argument caused the forked subagent to skip the skill's steps
and return a generic greeting instead of committing. Fork only works
reliably when the caller always supplies a rich task description as the
argument, which isn't true for these skills' common invocation.

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
This commit is contained in:
Jean-Pierre MartineauandClaude Haiku 4.5 committed 2026-07-12 11:35:45 -04:00
1 parent 56b7580e48
commit 534128ab84
2 files changed
+7

No files matched your search

+1
View File
@@ -1,6 +1,7 @@
---
name: verify
description: How to build, launch, and drive MeshTerm to verify changes at runtime.
model: haiku
---
# Verifying MeshTerm changes
+6
View File
@@ -7,6 +7,12 @@ everything heard is recorded to SQLite (`meshterm/persistence/`).
Run the tests with `python -m pytest -q`. Screens must stay readable at 72 columns.
## Git workflow
Commit directly on `main` — do not create topic branches unless explicitly asked (this
is a solo repo; branches just create divergence to merge back later). Commit freely, but
never `git push` unless explicitly asked in that moment.
## UX standards
These are binding. Every new screen, dialog, row, or hint follows them; when you touch an