Command reference¶
/mb is the single entrypoint for the whole toolkit — one command with subcommands, rather than
25 separate slash commands to remember. Every row below is available as /mb <subcommand>; a
handful also have short aliases (/start, /done, /plan, /discuss, /sdd, /work,
/verify, /config, /pipeline) that dispatch to the exact same underlying command.
| Subcommand | Purpose |
|---|---|
context (or empty) |
Collect and summarize current project context — status, checklist, active plan. |
start |
Extended session start: full context plus the active plan read in. |
search <query> |
Search core Memory Bank files. |
recall <query> |
Lexical recall over the session-memory log (session/) + notes/. |
recap <sid> |
Reconstruct a full progress.md entry from a session file that ended on an auto-capture stub. |
conflicts [--judge] |
Surface memory entries with high lexical overlap and an opposing/replacement assertion. |
consolidate [--apply] |
Fold old sessions into durable notes/ candidates; archive them verbatim. |
research <query> |
Graph-first multi-source research — codebase, memory, library docs, prior art, web. |
note <topic> |
Create a note. |
agree <add\|question\|list\|sync> |
Running registry of confirmed decisions (AGR-NNN, supersedes-chain); active list mirrored into CLAUDE.md/AGENTS.md. |
update |
Actualize core files from real code-state analysis (no session summary required). |
doctor |
Find and fix internal Memory Bank inconsistencies. |
tasks |
Show unfinished tasks. |
index |
Registry of all entries. |
done |
End the session — actualize, note, and append progress. |
plan <type> <topic> |
Create a plan (feature / fix / refactor / experiment). |
discuss <topic> |
5-phase requirements-elicitation interview producing an EARS-validated context/<topic>.md. |
sdd <topic> [--force] |
Create the spec triple specs/<topic>/{requirements,design,tasks}.md. |
config <init\|show\|validate\|path> |
Manage the project's pipeline.yaml. |
pipeline <list\|new\|use\|show\|path\|validate> |
Manage multiple named pipelines with different model routing per host. |
work [target] [--workflow NAME] [--range A-B] |
Execute a plan/spec through the composable implement→verify→(review→judge)→done loop. |
verify |
Verify plan/spec execution against the actual codebase — required before done when work followed a plan. |
map [focus] |
Scan the codebase, write STACK / ARCHITECTURE / CONVENTIONS / CONCERNS docs. |
upgrade [--check\|--force] |
Update a git-clone install of the skill from GitHub. |
compact [--dry-run\|--apply] |
Status-based archival decay for old completed plans and low-importance notes. |
import --project <path> [--apply] |
Bootstrap Memory Bank from Claude Code JSONL session transcripts. |
graph [--apply] [--cochange] [--questions] [--docs] |
Build/update the code graph (codebase/graph.json + god-nodes.md). |
wiki [--dry-run] |
Opt-in LLM layer: per-community codebase wiki + surprising cross-cutting connections. |
tags [--apply] [--auto-merge] |
Normalize frontmatter tags against a closed vocabulary. |
init [--minimal\|--full] |
Initialize Memory Bank in the current project. |
profile <subcommand> |
Manage rule profiles (role/stack/architecture/delivery-tuned rule sets). |
install [<clients>] |
Install Memory Bank for the project across one or more AI-agent clients. |
statusline [--force] |
Claude Code only — install the context-window statusline. |
help [subcommand] |
Print the router table, or the detailed section for one subcommand. |
deps [--install-hints] |
Check required/optional dependencies (python3, jq, git, …). |
idea <title> [HIGH\|MED\|LOW] |
Capture a new idea in backlog.md with an auto-generated I-NNN ID. |
idea-promote <I-NNN> <type> |
Promote a captured idea into a plan. |
adr <title> |
Capture an Architecture Decision Record with an auto-generated ADR-NNN ID. |
goal |
Scaffold goal.md + project.md, then validate the goal. |
analyze-task |
Auto-classify goal + git-diff scope into a Dynamic Flow route. |
flow <route> |
Explicitly select a Dynamic Flow route, skipping auto-classification. |
migrate-structure [--dry-run\|--apply] |
One-shot v3.0 → v3.1 structural migrator. |
Getting details on one subcommand¶
/mb help # router table (this page's source of truth)
/mb help work # full "### work" section: behavior, examples, underlying scripts
/mb help compact # full "### compact" section
/mb help <subcommand> extracts the exact ### <subcommand> block from commands/mb.md — the
same content this page condenses, kept as the single source of truth so the two never drift.
Related¶
- /mb work — the execution engine, documented in full.
- pipeline.yaml reference — the config
config/pipelinemanage. - Agents reference — the specialist subagents
work/map/verifydispatch.