memory-context-intelligence

Status: Plugin-scoped init-configuration surface completed in checked scope; /memory-context-intelligence:init now owns guided setup, /memory-context-intelligence:analysis remains the review surface, review and packet remain deferred, and packet/additional emission stays one-topic-per-artifact Current Version: 0.1.77 Plugin Package Version: 0.9.29 Session: d42465eb-30a7-4bc8-b9d6-03e52306e9a5 (2026-06-02)

What this plugin does

memory-context-intelligence is a Claude Code plugin that reads bounded work evidence from prepared memory traces and turns it into topic cards that help you spot recurring workflow, RULES, and operating-pattern issues worth reviewing.

In plain terms, this plugin is not for writing product features directly. It is for questions like:

Are we getting stuck on the same workflow problem repeatedly?
Is there a RULES gap that should be turned into a proposal?
Which governance or workflow topic should we investigate first to improve the next wave of work?

ภาพนี้เป็น visual identity ของ plugin ตัวนี้ — มันสื่อแนวคิดหลักของ plugin ว่าใช้ memory + context + evidence เพื่อช่วย review pattern การทำงาน ไม่ใช่เพื่อกระโดดไปแก้ product code ตรง ๆ

Current checked public surfaces:

/memory-context-intelligence:analysis
/memory-context-intelligence:init

Current checked behavior that matters:

the default mode is historical-first analysis, not current-session-only summary
trace_evidence remains the live promotion anchor
recall_evidence, durable_memory_context, and governance_context can strengthen wording and context, but they do not replace trace proof
when adaptive deep-analysis marks a top topic as requiring deeper review, the checked current skill contract now requires one bounded read-only deepening pass before the first response instead of silently skipping it
that deepening can use read-only subagent help plus optional web/external research support, but it still stays advisory-only and cannot replace local trace proof
if several advisory topics are deepened or compared, any later packet-derived or additional-stage output must still keep one selected topic per artifact and must split into separate per-topic artifacts instead of emitting a combined file
if no config file is loaded, the plugin can show a guided config helper and point the operator to /memory-context-intelligence:init instead of expecting raw arguments as the normal UX
the default config target is now the user-scope file ~/.claude/memory-context-intelligence.config.json

At a glance

Use this plugin when you want an evidence-first reflection tool that can:

scan historical work traces for repeated workflow/RULES issues
surface candidate topics before you decide what to improve next
keep trace evidence as the live anchor while still benefiting from recall, durable memory, and governance context
offer a guided setup path so the normal UX does not depend on raw config arguments

This plugin is especially useful after a work stretch when you want to understand:

what keeps repeating
what should become a proposal next
which issue is worth investigating first

When to use it

Use this plugin when you want Claude Code to review your past work patterns in an evidence-first way, for example:

after a stretch of work, when you want to see which blockers keep repeating
when you want candidate topics for improving RULES, workflow, or governance
when you want a broader historical review instead of a narrow current-session-only summary
when you want to limit source classes or historical depth through a config file without changing the core evidence model

This is not the right tool when you want to:

jump straight into fixing product code without caring about trace history
treat bare /analysis as the public command
rely on the plugin to auto-run in the background through hooks or monitors

Install and load this plugin

In checked scope, this plugin is currently distributed through a local marketplace named darkwingtm at the root TEMPLATE directory, and its marketplace entry points to:

./RULES/plugin/memory-context-intelligence

That means the current end-user flow is:

1) Prepare a local checkout of TEMPLATE

You need a local TEMPLATE directory on your machine, and its root must contain .claude-plugin/marketplace.json.

Portable placeholder:

<template-root> = the path to your local TEMPLATE directory

Example path shapes:

Linux / WSL: /home/<user>/work/TEMPLATE
Windows: C:\Users\<user>\work\TEMPLATE

2) Add the marketplace

Inside Claude Code, add the marketplace from the local path:

/plugin marketplace add <template-root>

If you prefer to point directly to the file:

/plugin marketplace add <template-root>/.claude-plugin/marketplace.json

3) Install the plugin

/plugin install memory-context-intelligence@darkwingtm

The checked Claude Code plugin docs say that direct installs like this use user scope by default.

If you want a different scope:

open /plugin
go to Discover or the plugin details view
choose user, project, or local

4) Reload plugins

/reload-plugins

5) Verify the install

The simplest verification path is:

open /plugin and confirm the plugin is installed and enabled
then call the slash surfaces below

First use workflow

Step 1 — guided setup

Inside the Claude Code session, run the setup surface first:

/memory-context-intelligence:init

That flow should use question/choice dialogs, default to Comprehensive default (Recommended), and write:

~/.claude/memory-context-intelligence.config.json

Step 2 — run the analysis surface

After setup, your normal review command remains:

/memory-context-intelligence:analysis

What you should expect

If usable evidence is available, you should get topic-card output such as:

Topic 1, Topic 2, Topic 3
each card explains what the topic is about
why it surfaced from evidence
what the before/after behavior looks like
that the result is still candidate only; advisory only; not approved yet
multiple topic cards are still advisory comparison output only; they are not a combined packet/additional artifact candidate
when the adaptive payload flags a top topic for deeper review, the first response may now include one bounded read-only deepening pass before the cards are finished, while still keeping the final result advisory-only and trace-anchored

At the end of the output, you should also see Next action options, such as:

choose a topic number
type a direct request
ask for deep thinking, websearch, or webfetch first

If no config file is loaded

If the current run does not load ~/.claude/memory-context-intelligence.config.json, you should see a Config helper that points you toward /memory-context-intelligence:init and asks operator-facing questions such as:

should the default scope stay broad historical-first or narrow down
which source classes should stay enabled
whether same-day widening should be allowed
whether the next config choice should be written now

Important boundary:

the helper is still advisory only inside /memory-context-intelligence:analysis
the dedicated write path now belongs to /memory-context-intelligence:init
/memory-context-intelligence:analysis remains the review surface

Default config file

The default config target is now:

~/.claude/memory-context-intelligence.config.json

The preferred way to create/update it is:

/memory-context-intelligence:init

Example shape:

{
  "analysis": {
    "scope_policy": {
      "default_scope_mode": "historical-comprehensive",
      "scope_day_shard": null,
      "scope_session_id": null,
      "scope_lookback_minutes": null
    },
    "source_policy": {
      "enabled_source_classes": [
        "trace_evidence",
        "recall_evidence",
        "durable_memory_context",
        "governance_context"
      ],
      "max_historical_shards": 10,
      "allow_same_day_widening": true
    }
  }
}

Current checked keys:

default_scope_mode
scope_day_shard
scope_session_id
scope_lookback_minutes
enabled_source_classes
max_historical_shards
allow_same_day_widening

Important behavior:

this config file is a late-bound analysis default, not semantic authority
scope_policy defines the stored default scope, but explicit runtime narrowing still wins
source_policy stays bounded to source classes, shard cap, and same-day widening
it does not create a fifth evidence class
it does not weaken the role of trace_evidence as the live promotion anchor
if you disable trace_evidence or leave only context-only sources in scope, the result must not be overclaimed as a promotable live candidate

Boundaries you should know

Current checked boundaries for this plugin:

the proved public analysis surface is /memory-context-intelligence:analysis
the setup/configuration surface is /memory-context-intelligence:init
bare /analysis is not current proved runtime behavior
/memory-context-intelligence:review and /memory-context-intelligence:packet are still deferred
bin/memory-context-intelligence is an internal implementation adapter, not the main user workflow
packet/additional-stage emission remains a single-topic-per-artifact flow; if several topics are ever carried forward, they must split into separate per-topic artifacts rather than one combined file
plugin-managed auto-flow is not currently claimed as proved behavior
this README does not claim external marketplace publication or broad production readiness

In plain terms: after installation, you should expect an analysis command that you run when you want a workflow-pattern review, plus a dedicated init/setup surface — not a plugin that silently runs itself all the time.

Troubleshooting

`/memory-context-intelligence:analysis` does not appear or does not work

Check in this order:

run /reload-plugins
open /plugin and confirm the plugin is still installed and enabled
confirm the installed source is memory-context-intelligence@darkwingtm
if the plugin was updated while this session was already running, restart the Claude Code session and retry

`memory-context-intelligence@darkwingtm` will not install

This is usually a marketplace-path problem.

Check that:

you added the correct <template-root> marketplace path
the root of that path contains .claude-plugin/marketplace.json
the marketplace entry for memory-context-intelligence still points to ./RULES/plugin/memory-context-intelligence

The result is `blocked`, `dormant`, or `no-topics`

That does not automatically mean the plugin is broken.

Typical meanings:

blocked = required analysis input or dependency is not ready
dormant = the available memory input is too stale
no-topics = there is not yet a repeated enough pattern to promote into a proposal

What to do next:

keep working until there is more trace history to analyze
adjust source policy or historical depth if needed
use a narrower scope intentionally if that better fits the review you want

The session is old and behavior does not match the latest source

The current checked contract includes an advisory warning named possible stale long-lived session.

If you see it:

restart the Claude Code session
run /memory-context-intelligence:analysis again

This is only a temporary diagnostic safeguard. It does not mean long-lived session mismatch is considered normal behavior.

Advanced notes

Evidence model at a glance

The current implemented evidence model uses 4 classes:

trace_evidence
recall_evidence
durable_memory_context
governance_context

Core rules:

trace_evidence remains the live anchor
source mix should stay visible when durable memory or governance context materially shaped a candidate
historical-first is the default, but explicit narrowing is still available when the operator wants it

Narrowing runs intentionally

If you want a narrower review slice, the current design still supports concepts such as:

day=YYYY-MM-DD
session=<id>
lookback=<minutes|hours>

But the checked public UX is intentionally built around the slash surface and guided helper first, not around making raw arguments the main normal path.

What this plugin is not

This plugin is not:

an automatic rule writer
a background monitor
proof that bare /analysis is publicly available
a shortcut for bypassing evidence gates
a replacement for real work traces

A better mental model is:

historical workflow reviewer
RULES / governance topic suggester
evidence-first reflection tool

Current truth summary

If you want the shortest version, remember these 5 steps:

add the local marketplace from <template-root>
install memory-context-intelligence@darkwingtm
run /reload-plugins
run /memory-context-intelligence:init
use /memory-context-intelligence:analysis

Image asset used by this README:

img/MCI.png

If your first run does not load a config file yet, that is fine — the plugin should guide you with the config helper and /memory-context-intelligence:init instead of forcing raw arguments as the normal starting UX.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

memory-context-intelligence

What this plugin does

At a glance

When to use it

Install and load this plugin

1) Prepare a local checkout of TEMPLATE

2) Add the marketplace

3) Install the plugin

4) Reload plugins

5) Verify the install

First use workflow

Step 1 — guided setup

Step 2 — run the analysis surface

What you should expect

If no config file is loaded

Default config file

Boundaries you should know

Troubleshooting

`/memory-context-intelligence:analysis` does not appear or does not work

`memory-context-intelligence@darkwingtm` will not install

The result is `blocked`, `dormant`, or `no-topics`

The session is old and behavior does not match the latest source

Advanced notes

Evidence model at a glance

Narrowing runs intentionally

What this plugin is not

Current truth summary

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

memory-context-intelligence

What this plugin does

At a glance

When to use it

Install and load this plugin

1) Prepare a local checkout of TEMPLATE

2) Add the marketplace

3) Install the plugin

4) Reload plugins

5) Verify the install

First use workflow

Step 1 — guided setup

Step 2 — run the analysis surface

What you should expect

If no config file is loaded

Default config file

Boundaries you should know

Troubleshooting

/memory-context-intelligence:analysis does not appear or does not work

memory-context-intelligence@darkwingtm will not install

The result is blocked, dormant, or no-topics

The session is old and behavior does not match the latest source

Advanced notes

Evidence model at a glance

Narrowing runs intentionally

What this plugin is not

Current truth summary

`/memory-context-intelligence:analysis` does not appear or does not work

`memory-context-intelligence@darkwingtm` will not install

The result is `blocked`, `dormant`, or `no-topics`