slop-janitor

Execute an approved .agent/goals plan through slop-janitor's goal runner, defaulting to .agent/goals/active when no explicit plan path is provided.

Turn a freeform idea, brief, or discussion into a durable sequential goal plan under .agent/goals/ for later execution by slop-janitor goals run.

Create a long-horizon meta-plan from a user brief, PRD, RFC, or dictated discussion. Use when the user wants to capture a multi-slice project in a durable parent artifact before creating child ExecPlans.

Create an ExecPlan from a locked refactor decision, PRD, RFC, or detailed problem statement, following the repo's PLANS.md. Use when the user asks for an exec plan, execution plan, or ExecPlan, or wants a decided refactor turned into a step-by-step plan.

Review the latest implemented work item or completed ExecPlan with fresh eyes, fix obvious issues immediately, rerun verification, and record the result. Prefer the new `.agent/work/` workflow, but preserve backward-compatible support for older `.agent/done/` ExecPlans.

Read an existing ExecPlan, deeply analyze every referenced file and code path, and rewrite the plan with concrete, code-grounded improvements. Prefer work-item plans under `.agent/work/`, while preserving backward-compatible support for older singleton plan files.

Improves an existing ExecPlan with a two-wave subagent workflow: first specialized reviewers audit factual accuracy, adjacent code, design quality, validation, and self-containment, then closure reviewers hunt residual gaps after a provisional rewrite before the parent produces the final plan. Use when the user asks to improve an execplan with subagents, audit a plan using parallel reviewers, strengthen an execplan via delegation, or says "execplan-improve-subagents".

Search a repo for the top 3-5 materially different refactor opportunities, record the evidence for each, and create a work item for later selection and planning. Use when the user wants the best refactor, highest-leverage cleanup, architectural simplification, boundary extraction, duplication removal, or a shortlist of strong refactor options before choosing one.

Execute an ExecPlan from a work item under `.agent/work/` or, when necessary, from the older singleton `.agent/execplan-pending.md` flow. Use when the user asks to implement a plan, execute the pending plan, or resume blocked implementation work.

Run a two-wave subagent review-and-fix pass on the most recently implemented ExecPlan or a user-specified completed ExecPlan. Specialized reviewer agents inspect contract correctness, regression surface, failure modes, state lifecycle, ownership, validation, and simplicity boundaries, then closure agents audit the synthesized fix plan. The parent agent implements the selected fixes, reruns verification, and reports the result. Use when you want post-implementation review after an ExecPlan lands, want independent review lanes that report back to the main agent, or want an Ousterhout-focused fresh-eyes audit that ends in concrete fixes.

Read a refactor candidate shortlist, pressure-test the leading options with cheap repo evidence, and lock the final refactor choice in a work item before planning. Use when the user wants the best refactor chosen from a shortlist, wants the current favorite challenged, or wants a final decision before creating an ExecPlan.