name	review-boundary
description	Review AgentM code for boundary isolation, pluggability-axis violations, atom contract breaches, and design-pattern intrusion. Use when asked to run a boundary review over a path, git range, current branch, or staged changes; also trigger on "review-boundary", "boundary reviewer", "boundary audit", "pluggability review", or "§11 contract review".

review-boundary

Audit AgentM code for boundary isolation, pluggability-axis violations, and design-pattern intrusion.

Treat the user's text after the skill name as the review selector:

Empty / omitted: review src/agentm/.
A path or glob, such as src/agentm/gateway/ or contrib/extensions/llmharness/: review that subtree.
A git range, such as HEAD~5..HEAD or main..HEAD: review files touched in that range.
--branch: review files changed on the current branch vs main (git diff --name-only main...HEAD).
--staged: review currently staged files.

If the selector is ambiguous, ask the user before reviewing.

Resolve the file list from the selector.
Capture git context: current branch, base ref when relevant, and whether there are uncommitted changes.
Read the relevant code and design context. Prefer:
- .Codex/designs/pluggable-architecture.md
- AGENTS.md
- nearby design docs under .Codex/designs/
- CONTEXT.md when terminology is unclear
Check for:
- direct atom imports from agentm.core.runtime.*, core._internal, or other forbidden kernel internals;
- atom-to-atom imports or scenario-specific logic leaking into agentm.core;
- bypasses of ExtensionAPI services for stateful subsystems;
- changes that weaken §11 extension validation;
- pluggability-axis behavior embedded in the substrate instead of registered through atom protocols;
- design docs or .Codex/index.yaml left stale after concept changes.
Report findings first, ordered by severity, with concrete file and line references.

Use code-review shape:

## Findings
- [block|major|minor] path:line - Issue and consequence.

## Open Questions
- ...

## Summary
- Reviewed scope and residual risk.

If any finding is block, end by saying merging is not advised until the blockers are resolved.

review-boundary

Audit AgentM code for boundary isolation, pluggability-axis violations, and design-pattern intrusion.

Input

Treat the user's text after the skill name as the review selector:

Empty / omitted: review src/agentm/.

A path or glob, such as src/agentm/gateway/ or contrib/extensions/llmharness/: review that subtree.

A git range, such as HEAD~5..HEAD or main..HEAD: review files touched in that range.

--branch: review files changed on the current branch vs main (git diff --name-only main...HEAD).

--staged: review currently staged files.

If the selector is ambiguous, ask the user before reviewing.

Review Procedure

Resolve the file list from the selector.

Capture git context: current branch, base ref when relevant, and whether there are uncommitted changes.

Read the relevant code and design context. Prefer:

Check for:

direct atom imports from agentm.core.runtime.*, core._internal, or other forbidden kernel internals;
atom-to-atom imports or scenario-specific logic leaking into agentm.core;
bypasses of ExtensionAPI services for stateful subsystems;
changes that weaken §11 extension validation;
pluggability-axis behavior embedded in the substrate instead of registered through atom protocols;
design docs or .Codex/index.yaml left stale after concept changes.

Report findings first, ordered by severity, with concrete file and line references.

Output

Use code-review shape:

## Findings - [block|major|minor] path:line - Issue and consequence. ## Open Questions - ... ## Summary - Reviewed scope and residual risk.

If any finding is block, end by saying merging is not advised until the blockers are resolved.

Plus depuis ce dépôt