agent-skills

Run Cursor Agent headlessly and delegate work to another Cursor agent over the CLI. Use when the user asks to call, ask, invoke, run, or delegate to Cursor (e.g. "call cursor", "ask cursor", "invoke cursor agent"); when automating Cursor from scripts or agent-to-agent flows; or when the user mentions `agent -p`, headless Cursor CLI, or `cursor-agent.mjs`.

Hand off work in progress or finished to another agent for continuation or review. Produce a structured summary with enough background context, what was worked on, how, why, files touched, and what remains. Trigger when the user says "hand off this work", "prepare a handoff", "document what was done for the next agent", or ending a session another agent should pick up.

Review recently modified code for clarity, consistency, and maintainability while preserving exact functionality. Audit adherence to project conventions and industry best practices. Trigger when the user wants a code quality review, readability audit, maintainability review, or behavior-preserving refinement suggestions on a diff or implementation.

Execute an approved plan or spec document phase-by-phase, writing robust idiomatic code that follows codebase patterns. Trigger when the user says "implement this plan", "execute the spec", "work through the phases", or a plan in `dev/plans/` is approved and ready for implementation.

Review a given implementation critically and adversarially against its plan or spec. Look for antipatterns, red flags, bugs, unnecessary complexity, plan drift, and missing tests. Trigger when the user says "review this implementation", "review these changes", "review this branch", "adversarial review", "challenge these changes", or wants code scrutinized before merging.

Create a scoped, code-backed implementation plan from a todo, spec, issue, review notes, or raw user instructions. Use when the user asks to convert requirements into a proper plan, phased implementation plan, executor handoff plan, or reviewable planning artifact before coding.

Survey any codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute. Strictly read-only on source code — never implements, fixes, or refactors anything itself. Use when asked to audit a codebase, find improvement opportunities (bugs, security, performance, test coverage, tech debt, migrations, DX), suggest features or where to take the project next (roadmap, product direction), or generate handoff plans for another agent to implement.

Research and define codebase issues before implementation planning. Use when the user provides an issue statement, bug report, Jira ticket, vague problem, proposed solution, runtime symptom, or design concern and wants Codex to inspect the current codebase, validate whether the problem exists, diagnose likely causes, compare solution directions, and produce an evidence-backed problem definition. Do not use when the user asks for a step-by-step implementation plan, direct implementation, or code review of an existing diff.

Evaluate, analyze, and systematically react to an adversarial code review report. Decide on the action for each finding (Implement, Adapt, Decline), provide clear justifications, and define the plan to implement the accepted changes.

Interview the user to extract knowledge from their head and synthesize it into a structured document. Use this skill when the user wants help thinking through an idea, designing a feature, specifying a task, or articulating something they haven't written down yet. Trigger when the user says things like "interview me about", "ask me questions about", "help me think through", "I need to spec out", "I have an idea for", "let's flesh out", or any variation where they want to be interrogated rather than do the writing themselves. Also use when the user has a vague concept and needs help turning it into a concrete artifact (spec, design doc, brief, plan, etc.) through conversation.

Review a spec document against codebase reality, identifying gaps and ensuring sound, robust implementations.

Clarify requirements before implementing. Do not use automatically, only when invoked explicitly.