Make Compass reviews reduction-first - #280
Conversation
ariobarin
left a comment
ariobarin
left a comment
There was a problem hiding this comment.
Current-head review at 73b869cce7c117653bd6fa9b91d077aeff13b0d0: no blocking findings. The reduction-first rewrite preserves audience, behavior, evidence, authority, wiring verification, long-running control, and taste-boundary requirements while removing duplicate doctrine. portable checks run 1057 completed successfully, with no existing review comments or unresolved threads. Ready for merge under the user's explicit authority.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 73b869cce7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| For long-running guidance, verify that one logical principal authors control | ||
| state, delegates receive reviewed assignments and return evidence, and a fresh | ||
| context can resume from anchors plus a checkpoint. |
There was a problem hiding this comment.
When this program is used to review long-running guidance, this reduced checklist no longer asks reviewers to verify that planning authority is explicit before production mutation; it only checks authorship, evidence, and resume behavior, and the later public-mutation sentence does not cover local production edits. That lets a workflow or skill pass this review while allowing implementation to begin from a planning-only assignment, which is a failure mode this repo tells us to flag. Re-add the production-mutation authority check in the long-running guidance review.
AGENTS.md reference: AGENTS.md:L96-L96
Useful? React with 👍 / 👎.
| - Does the surface shape judgment instead of enumerating every thought? | ||
| - Does it overlap another role, or merely compose with it? | ||
| - Would a project or carried route preserve value at lower global cost? | ||
| - Does model and effort routing match the current dated profile? |
There was a problem hiding this comment.
When reviewing a skill or agent that changes routing, this replacement asks only whether routes match the “current dated profile” but no longer points to local-docs/model-calibration.md or says to update calibration and runtime files together. Because the active Sol, Luna, Terra, GLM, and effort policy is dated and stored outside the runtime files, following this workflow alone can leave a role routed by memory or stale policy instead of current evidence. Restore the calibration source in this check.
AGENTS.md reference: AGENTS.md:L98-L98
Useful? React with 👍 / 👎.
1 participant
Compass already values reduction, but the review workflow repeated doctrine without a compact editing test. This distills the audit into five moves and removes 41 net lines from that workflow while preserving verification and authority boundaries.