ai_tools

Parses the repo and conversation to produce a self-contained markdown advisory brief for an external AI (no codebase access) to evaluate options, recommend a path, and return structured deliverables. Use when comparing architecture, vendors, stacks, or tradeoffs before implementation — e.g. "evaluate DB options", "compare auth providers", "should we use X or Y".

Creates implementation plans following PLANNING_RULES. Use when the user asks to create a plan, write an implementation plan, or plan out work. Ensures Overview, Happy Flow, Manual Verification, Update Runbooks (audit docs/runbooks/), parallel delegation, asset storage, and UI screenshots. Loads phase-specific reference files before each gate.

Produces a concise end-of-session handoff in `handoff.md` summarizing what was done, what remains (with the last to-do you were working on), the PR link, and the key plan/notes files to read next. Use at the end of a work session or before switching agents.

Investigates failing GitHub PR CI checks, reproduces failures locally, fixes scoped issues, commits and pushes, and summarizes changes. Use when CI is red, GitHub Actions or checks fail on a PR, or the user asks to fix CI/build/ lint/test failures for a branch or pull request.

Diagnoses a microservice or pipeline service and produces a refactoring plan focused on modularity, test coverage, lint/type health, runbooks, and service READMEs. Use when the user asks to refactor a service, pipeline, microservice, worker, batch job, or study service and wants a plan before implementation. Slash-only.

Converts user-provided text or image content into LaTeX-only output using inline $...$ and display $$...$$ delimiters with no bracket delimiters and no additional commentary. Use when the user asks to write content in LaTeX with strict output-only formatting.

Reviews plans, proposals, and code changes for unnecessary complexity, unverified assumptions, and premature abstraction. Use when the user asks to simplify a plan, challenge a proposal, narrow scope before implementation, pressure-test whether a design is overengineered, or review a diff for avoidable complexity. Slash-only.

Implements an existing implementation plan to completion, verifies the result, creates a pull request using the local create-pr skill, and returns the PR URL. Use only when the user explicitly asks to execute a plan end-to-end, open a PR, and provide the link.

Performs thorough code reviews using the collaborative review checklist. Use when the user asks for a comprehensive code review, collaborative review, thorough review, or structured review with per-file analysis, cross-cutting concerns, and risk assessment.

Probes the user's understanding of an engineering task or design as an experienced engineering manager. Use when the user wants to explain a feature, defend a design, pressure-test requirements, challenge assumptions, rehearse system design, or review technical and non-functional trade-offs.

Tests and deepens the user's understanding of a topic as an expert professor. Use when the user wants to be quizzed, explain something at different levels, increase difficulty over time, validate conceptual depth, or get critical feedback on understanding across engineering, math, AI, or other technical topics.

Drafts or refines PR descriptions following HOW_TO_WRITE_PR rules. Use when creating a PR, writing a PR description, or when the user asks for help with a pull request.

Applies code-security (Semgrep) and security-best-practices (OpenAI) to the current task. Use when the user asks for a security review, secure code guidance, or vulnerability check. Instructs the agent to run both skills.

At end of conversation, analyzes the exchange to infer user preferences and suggests additions to docs/RULES.md. Use when the user asks what preferences they could add to their rules, or wants to capture practices from the conversation.

For Python-fluent developers. Explains code in other languages (especially TypeScript) through a Python lens—teaches concepts first (e.g. what is a hook?), then execution/model and idioms, then translation or line-by-line mapping to Python. Use when the user references a code chunk or asks about code in a non-Python language and wants to understand it, learn the language, or get an idiomatic Python equivalent.

Reviews current work using a selected persona from agents/personas. Use when the user asks for persona-based review, expert-lens critique, or domain-specific code assessment.

Reviews current work against task-instruction rules in agents/task_instructions/rules. Use when the user asks for rules-based compliance checks or policy-aligned review feedback.