Where this fits

A developer hands your agent a rough plan: a few bullet points in chat, a half-completed plan task, or a paste of "we should build X, maybe with Y." Your agent could start coding, but the missing scope, acceptance criteria, and test strategy will surface later as rework. gza-plan-improve sits in that gap: it runs a deliberate question loop, closes the highest-leverage gaps, and rewrites the draft into an implementation-ready plan before a single line of code is written.

Why agents benefit

• Closes ambiguity without guessing. The skill uses AskUserQuestion to confirm assumptions explicitly, so the agent never invents constraints the user didn't state.
• Produces a rewritten plan, not a score. The output is a concrete, refreshed plan with sharper scope, acceptance criteria, sequencing, risks, and test strategy — ready to feed into the next step.
• Accepts either a full prefixed task ID (gza-1234) or raw pasted draft text, so it works whether the user works inside a gza project or just pastes notes.
• Prioritizes questions by leverage. It identifies the biggest blind spots first instead of running a generic checklist, keeping the interaction short.
• Logs the refinement with gza log/show commands, so decisions stay traceable for later audit.

Practical scenarios

Incomplete plan task after an interrupted session

Your agent's previous run stopped mid-planning. The user points at task gza-0881. The skill reads the existing plan, spots missing acceptance criteria and unsequenced steps, asks three targeted questions, and rewrites the task so implementation resumes cleanly.

Rough idea pasted from Slack

A stakeholder pastes "we need rate limiting on the API, probably with Redis, and maybe per-tenant." This skill extracts the weak assumptions (which limits? which Redis patterns? what happens on overflow?), asks about edge cases, and returns a structured plan the developer can approve.

Migrating a legacy service with unclear rollback

The draft says "move to the new auth service." The skill surfaces missing rollback steps, risk ratings for each migration phase, and a test matrix — turning a one-liner into a deployable plan.

Add it to your agent workflow

Install the skill:

npx skills add mhawthorne/gza --skill gza-plan-improve

Then invoke it from your agent with either a gza task ID or pasted draft text, plus optional constraints. For example: "Refine plan task gza-1234; the weak areas are sequencing and the acceptance criteria for the batch job."

Compare before adopting

Before you add this skill, compare it against alternatives on four axes: quality signals (does the rewritten plan survive implementation review?), maintenance freshness (check last commit and issue responses), alternatives (other plan-enhancement skills or a manual prompt you already use), and workflow fit (does your agent already interact via AskUserQuestion? Does it have access to a gza project?).

Why it is worth tracking

At 11 stars, this is early-stage, but the pattern it formalizes — a question loop that improves plans rather than scoring them — is genuinely useful for agents that plan before they act. Watch whether the author sustains momentum (releases, issue triage) and whether the skill broadens beyond gza-native task IDs. Evaluate it whenever your agent's plans routinely hit rework due to ambiguous requirements. For a broader view of plan-refinement tools, check the skills directory for related entries and community signals.