using-mozg-pipelines

name	using-mozg-pipelines
description	ALWAYS invoke when user asks which mozg skill/plugin to use, says 'what plugins do I have', 'which pipeline fits', 'route this'. Maps task phrases to skills across mz-dev-base, mz-dev-pipe, mz-dev-git, mz-research-pipe, mz-memory, mz-biz-outreach, mz-job-outreach, mz-creative, mz-gov.
argument-hint	[task description]
model	haiku
allowed-tools	Read, AskUserQuestion

Overview

This is a routing map, not a discipline skill. It is loaded into every new session via the mz-dev-base SessionStart hook and tells Claude which concrete skill to invoke for a given user task phrase. When the user describes work in natural language, consult this map first, then delegate to the matching skill. Do not attempt the work under this skill's own identity.

When to Use

Invoke when the user:

• asks "which skill/plugin/pipeline do I use for X"
• says "what plugins do I have" or "what can mozg do"
• asks to "route this task" or "pick the right pipeline"
• describes a task without naming a skill and you need to dispatch

When NOT to use

• The user already named a specific skill ("run audit", "use build") — invoke it directly.
• The user is mid-phase inside another skill — stay inside that skill.
• The user asked a direct question answerable without routing (code read, quick fix) — answer it.
• The task matches exactly one obvious skill from context — invoke it without re-citing the map.

Core Process

1. When the user states a task, map the phrase against the Techniques table BEFORE drafting an answer.
2. If the phrase matches multiple skills, list all candidates with the distinguishing condition, ask which fits, or pick the most specific.
3. If the phrase matches no entry, say so explicitly — do not invent a routing.
4. Prefer the most specific skill over a general one (e.g., audit over debug certainty:low for security review).
5. Do not paraphrase the user's task to fit a different skill's territory.
6. After routing, invoke the target skill directly. Do not re-explain the routing decision to the user unless they asked.

Techniques

Routing table. Phrases are indicative, not exact matches.

Task phrase	Skill	Plugin
"build a feature", "implement X", "add capability"	build	mz-dev-pipe
"fix this bug", "debug", "why is this broken"	debug	mz-dev-pipe
"security review", "audit for vulns", "threat model"	audit	mz-dev-pipe
"verify this works", "prove correctness", "check behavior"	verify	mz-dev-pipe
"polish this", "clean up", "finalize"	polish	mz-dev-pipe
"decomment", "strip AI comments", "remove robotic comments", "clean LLM comments from code"	decomment	mz-dev-pipe
"optimize", "make it faster", "profile hotspot"	optimize	mz-dev-pipe
"refactor", "remove dead code", "remove unused imports", "reduce complexity"	cleanup	mz-dev-pipe
"explain this code", "what does this do"	explain	mz-research-pipe
"investigate X", "dig into", "find root cause", "is X actually doing Y", "verify hypothesis"	debug certainty:low	mz-dev-pipe
"blast radius", "what breaks if I change X", "impact analysis"	audit depth:deep scope:branch (auto-invokes blast-radius)	mz-dev-pipe
"deep audit", "pre-PR audit", "ship audit", "before publishing", "before I open a PR"	audit depth:deep	mz-dev-pipe
"research topic deeply", "survey the field"	deep-research	mz-research-pipe
"synthesize past research", "combine findings", "consolidate prior work"	combine	mz-research-pipe
"translate this", "localize", "translate to "	translate	mz-research-pipe
"bootstrap rules", "init project rules"	init-rules	mz-dev-base
"review my branch", "what changed locally"	review-branch	mz-dev-git
"review this PR", "what's wrong with this PR"	github-review-pr	mz-dev-git
"scan open PRs", "triage PRs"	github-scan-prs	mz-dev-git
"help me author a new skill", "write a SKILL.md", "construct a skill"	construct-skill	mz-dev-base
"which skill fits", "route this", "what plugins do I have"	using-mozg-pipelines	mz-dev-base
"find leads", "customers matching X", "outreach list", "research targets", "find contacts at ", "decision-makers at X", "who do I write to"	outreach-research	mz-biz-outreach
"draft outreach letters from this card", "write letters from ", "enrich this card", "deepen research on ", "go deeper on "	outreach-enrich-company	mz-biz-outreach
"log action on ", "update card with ", "record interaction on ", "I just , add to "	outreach-update-card	mz-biz-outreach
"find a job", "search for vacancies", "job hunt", "match jobs to my CV"	job-search	mz-job-outreach
"find a recruiter", "hiring manager for X", "careers email for X", "who do I apply to"	job-recruiter-info	mz-job-outreach
"brainstorm", "generate ideas", "creative options"	brainstorm	mz-creative
"should I do X or Y", "record this decision", "write an ADR", "RFD for X", "is this decision sound"	govern	mz-gov
"install governance policy", "init governance", "set up decision records"	gov-init	mz-gov
"review our ADRs", "audit decision records", "are our design docs sound"	gov-review	mz-gov

mz-memory contributes SessionStart/SessionEnd memory hooks and has no user-facing skills; it is referenced by agents that need persistent state.

Common Rationalizations

N/A — informational routing skill, not a discipline skill.

Red Flags

• You answered a routing-type question by attempting the task directly without checking the routing table.
• You routed to a general skill (debug certainty:low) when a specific one (audit, audit depth:deep) exists for the phrase.
• You paraphrased the user's task into a different skill's territory to justify a skill you prefer.

Verification

To confirm the router was loaded into this session:

• Grep the initial system/context for the string using-mozg-pipelines or Mozg pipelines routing map.
• Ask Claude to recite the first column header of the routing table (Task phrase | Skill | Plugin).
• If neither is present, the SessionStart hook did not fire — re-enable the mz-dev-base plugin or run the session-start script manually.