| name | reviewing-framework-context-file-changes |
| description | Review changes in framework context files for conciseness, minimality, task fit, framework integration, ambiguity, actionability, verifiability, contradictions, scope clarity, and non-English text. Use when Codex reviews diffs or changed files in `.agents/`, `src/project-baseline.md`, `src/conventions/`, `src/roles/`, or framework templates. |
Review framework context file changes
Scope
Treat these as framework context files:
.agents/**
src/**
AGENTS.md
skills/installing-framework/**
Input
- diff or changed files
- original task
If the original task is not explicit:
- use
$finding-codex-session to locate recent session candidates
- read only material turns from candidate sessions
- show the likely task or concise alternatives to the user
- continue only after the user confirms the task
- if no likely task is found, ask the user for it
Checks
Check that changes are:
- written in English unless the file explicitly requires another language
- effective for the stated task
- integrated into the framework
- concise
- non-duplicating
- unambiguous
- actionable by an agent
- verifiable from files or output
- non-contradictory to existing framework files
- clear in scope of applicability
- sufficient to detect under-typed data structures when changed code represents domain states, variants, or semantic subgroups with correlated nullable fields
Integration
Verify:
- names, paths, and links are correct
- new files are reachable from existing entry points when needed
- old files were updated when new files replace or extend them
- no orphan files were introduced
Output
Return:
- verdict:
ok | ok_with_notes | needs_fix
- findings list
For each finding include:
- severity:
major | minor
- file
- issue
- why it is a problem
- minimal fix, including suggested English wording for non-English text
Review rule
Prefer deletion, shortening, reuse, and direct fixes over expansion.
Treat ambiguity, non-actionability, non-verifiability, contradiction, unclear scope, missing integration, and unintended non-English text as defects.
For non-English text findings, suggest an English rewrite that preserves intent and maximizes inference impact.
Do not praise.
Report only real issues.