| name | using-factor-harness |
| description | Orient across the four factor-harness skills (using-factor-harness, factor-architect, factor-scan, factor-dedup) and route to the right one. factor-harness is the small skill set for two recurring code-quality activities: architecture-aware review/refactor and cross-implementation deduplication. Each workflow skill produces an inline spec that hands off to Superpowers writing-plans for the actual code changes. Use this skill when the user mentions architecture review, refactor opportunities, code scanning, bug hunting, or duplicated implementations — even if they don't name a specific factor-harness skill. |
| when_to_use | When the user starts a session that mentions architecture, refactoring, code review, bug hunting, quality checks, or duplicate implementations; when the user asks "what factor-harness skills exist?"; before invoking any specific factor-* skill if the right choice is unclear. |
| allowed-tools | Read |
using-factor-harness — orient across the bundle
factor-harness provides three workflow skills for recurring
code-quality activities. Each produces an inline spec that hands
off to Superpowers writing-plans / executing-plans for the
actual refactor. factor-harness does the thinking; Superpowers
does the doing.
See ../_conventions.md for shared
conventions across the four skills.
The three workflow skills
| Skill | Trigger | Output |
|---|
factor-architect | A spec/plan was just produced (pre-execution) OR features have shipped (post-execution) and you want architectural review | Architectural review + refactor spec |
factor-scan | You want a broad bug + quality + architectural-smell sweep over a code area | Per-finding inline spec |
factor-dedup | You have N implementations doing similar things and want to consolidate | Comparison + unified-design spec (or documented divergence) |
Routing — when to use which
| If the user says... | Reach for... |
|---|
| "I just got a plan back from /writing-plans, can you review it?" | factor-architect |
| "I just shipped feature X, anything to refactor?" | factor-architect |
| "Does this match our patterns?" / "Is this architecturally sound?" | factor-architect |
| "Scan this directory for bugs and quality issues" | factor-scan |
| "Audit these files" / "Review for problems" / "Code health check" | factor-scan |
| "What's wrong with this code?" / "Any issues here?" | factor-scan |
| "These N files look duplicated" | factor-dedup |
| "Consolidate these implementations" / "DRY this up" | factor-dedup |
| "5 commands doing the same thing" | factor-dedup |
When the right choice is genuinely unclear, ask one clarifying
question rather than guessing — the wrong skill produces a wrong
shape of output.
Two adjacent tools
factor-harness is one of three plugins in the user's tool family.
Each has a clear lane:
- project-harness
— project-level lifecycle. Track what work needs doing
(PROJECTS.md trunk, per-project files). Orient: what's next,
what's in progress.
- Superpowers — execution discipline. Brainstorming, writing
plans, executing plans, code review.
- factor-harness (this plugin) — quality activities sitting
between project-level planning and per-task execution. Takes
in a plan or shipped code; produces a spec the user feeds back
into Superpowers.
The three plugins compose: project-harness might surface an
architectural concern as a refined project; factor-architect
runs the review and produces a refactor spec; Superpowers'
writing-plans turns that into an executable plan;
executing-plans runs the plan; the loop closes.
Core conventions (every factor-* skill follows these)
These are condensed pointers; full text is in
../_conventions.md.
- Pure workflow shape. No trunk file. No managed-item IDs.
No audit cycle. Skills are runbooks invoked when there's
specific work to do.
- Per-finding interactive UX. Every finding walks
accept / modify / skip / mark-intentional. The user's
judgment gets captured in the spec.
- Inline output. Skills produce findings and specs in
chat. No persistent artifacts. The user moves spec content
to Superpowers manually.
- Read-only. No factor-harness skill modifies code.
Execution is Superpowers' job.
- Three subagents —
factor-scanner (parallel scan),
factor-comparator (per-implementation read),
factor-researcher (external pattern research with web
access). Each calibrated for one shape of read; dispatched
in parallel where the work allows.
Iron laws (one per workflow skill)
Each workflow skill's SKILL.md states an inviolable discipline up
front:
factor-architect — Existing precedent is not automatically
good precedent. Refactor both when the precedent is poor.
factor-scan — Severity defaults conservative. The user
lifts severity, not the skill.
factor-dedup — "These look duplicated" is a hypothesis,
not a fact. The comparators may surface intentional divergence
and the right answer is "leave them separate".
What this skill does NOT do
- Does not invoke any of the three workflow skills. It teaches;
the user invokes (or references this routing table).
- Does not modify code, files, or repo state.
- Does not maintain state across sessions. Each workflow skill is
invoked freshly when its trigger fires.
See also