~/.codex/changes.md              # global memory
<project>/.codex/changes.md      # project memory

api, backend, codex, components, config, database, docker, docs, frontend,
git, i18n, json, mcp, migration, mongo, naming, opensearch, performance,
security, styles, tests

npx -y --package github:formonkey/knowledge-memory-mcp#main knowledge-memory-mcp

node /path/to/knowledge-memory-mcp/src/index.js

[mcp_servers.knowledge_memory]
command = "npx"
args = [
  "-y",
  "--package",
  "github:formonkey/knowledge-memory-mcp#main",
  "knowledge-memory-mcp"
]
enabled = true
startup_timeout_sec = 20
tool_timeout_sec = 60
default_tools_approval_mode = "auto"

[mcp_servers.knowledge_memory]
command = "node"
args = [
  "/absolute/path/to/knowledge-memory-mcp/src/index.js"
]
enabled = true
startup_timeout_sec = 20
tool_timeout_sec = 60
default_tools_approval_mode = "auto"

[mcp_servers.knowledge_memory.env]
CHANGES_MEMORY_GLOBAL_PATH = "/Users/nigma/.codex/changes.md"

~/.codex/config.toml                                # one global MCP server config
<project>/.codex/agents/knowledge-reviewer.toml     # optional read-only reviewer agent
<project>/.agents/skills/knowledge-memory-review/SKILL.md  # optional reviewer skill
<project>/AGENTS.md                                 # optional project-wide agent rules

~/.codex/config.toml

[mcp_servers.knowledge_memory]
command = "npx"
args = [
  "-y",
  "--package",
  "github:formonkey/knowledge-memory-mcp#main",
  "knowledge-memory-mcp"
]
enabled = true
startup_timeout_sec = 20
tool_timeout_sec = 60
default_tools_approval_mode = "auto"

<project>/.codex/agents/knowledge-reviewer.toml

name = "knowledge_reviewer"
description = "Read-only reviewer that checks code against knowledge_memory before and after implementation."
model = "gpt-5.5"
model_reasoning_effort = "high"
sandbox_mode = "read-only"

developer_instructions = """
You are a read-only knowledge reviewer for this repository.

Required MCP server:
- knowledge_memory

Available knowledge_memory tools:
- list_change_index
- list_tag_catalog
- get_change
- search_changes
- get_relevant_changes

Guardrails:
- Do not edit files.
- Do not run write, format, migration, install, or destructive commands.
- Do not call add_local or add_global.
- Do not save memory. If a new reusable rule is found, propose it to the main agent and wait for explicit user confirmation.
- Prefer compact reads: call list_change_index first, then retrieve only relevant entries with get_change, search_changes, or get_relevant_changes.
- When reviewing multiple projects, pass projectPath to knowledge_memory tool calls.

Review workflow:
1. Call list_change_index with projectPath when available.
2. Use tags from the index to decide which entries matter.
3. Retrieve only the relevant entries.
4. Review the current task or diff against those entries.
5. Report findings first, ordered by severity, with file and line references when available.
6. Include an explicit "Memory Checks" section listing which memory ids were applied or saying that no relevant entries were found.

Output format:
- Findings
- Memory Checks
- Open Questions
- Suggested Memory To Save, only if applicable and only as a proposal
"""

<project>/.agents/skills/knowledge-memory-review/SKILL.md

---
name: knowledge-memory-review
description: Review a task, plan, or diff against knowledge_memory using the compact memory index first.
---

# Knowledge Memory Review

Use this skill when reviewing implementation plans, diffs, bug fixes, or refactors against saved project and global memory.

## Required MCP Server

- `knowledge_memory`

## Read-Only Tools

- `list_change_index`
- `list_tag_catalog`
- `get_change`
- `search_changes`
- `get_relevant_changes`

## Guardrails

- Do not edit files.
- Do not run write, format, migration, install, or destructive commands.
- Do not call `add_local` or `add_global`.
- Do not save memory directly.
- If a new reusable rule should be saved, propose it and wait for explicit user confirmation.
- Prefer compact reads: call `list_change_index` first, then retrieve only the entries that look relevant.
- When reviewing several projects in one conversation, pass `projectPath` to memory tool calls.

## Workflow

1. Call `list_change_index` with `projectPath` when available.
2. Select candidate entries by tags, paths, title, and summary.
3. Retrieve only relevant entries with `get_change`, `search_changes`, or `get_relevant_changes`.
4. Review the task, plan, or diff against those entries.
5. Report findings first, ordered by severity, with file and line references when available.
6. Include a `Memory Checks` section listing the memory ids applied, or state that no relevant entries were found.
7. Include `Suggested Memory To Save` only when there is a reusable learning, and only as a proposal.

Before implementing, reviewing, or fixing code, use the `knowledge_memory` MCP.

Start with `list_change_index` to inspect the compact index. Use tags to decide which entries are relevant, then call `get_change`, `search_changes`, or `get_relevant_changes` only for those entries.

Use `add_local` only after explicit user confirmation for project-specific learnings.
Use `add_global` only after explicit user confirmation for cross-project learnings.
Every saved memory entry must include 2-5 tags from `list_tag_catalog`.

When a conversation touches multiple projects, pass `projectPath` to select the correct local memory file.

Ask knowledge_reviewer to inspect memory for this repo and review the planned change before I implement it.

Ask knowledge_reviewer to review this diff against knowledge_memory and report any repeated mistakes or missing conventions.

Save this as local memory: React components in this repo must use PascalCase file names and include the Component suffix. Tags: components, naming, frontend.

Save this as global memory: Always check list_change_index before reading full memory entries. Tags: codex, mcp, performance.

Before implementing or reviewing changes, call `list_change_index` on the `knowledge_memory` MCP server to inspect the compact index. Then call `get_change`, `search_changes`, or `get_relevant_changes` only for entries that look relevant.

When a conversation touches multiple projects, pass `projectPath` in tool calls to select the correct local memory.

Do not store memory on your own initiative. If the user confirms that a learning should be saved, use `add_local` for project-specific criteria and `add_global` only for criteria that apply across projects. Always include 2-5 tags from the catalog.