agent

Prepare and execute high-quality Conventional Commits from working tree changes with repository-aware commit scoping (root repo and submodules) and atomic commit planning. Use when asked to commit, split commits, draft commit messages, or organize staged/unstaged changes into safe commit units.

Design and implement code changes end-to-end, from requirements clarification through implementation, validation, and documentation updates. Use when writing new code, extending features, fixing bugs, or making structured code changes where correctness, maintainability, and clear tradeoffs matter.

Add or improve code and repository documentation so behavior, constraints, and usage are clear and current. Use when asked to write/update docs, docstrings, inline rationale comments, architecture notes, or change-related documentation for code updates.

Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

Install skills from skills.sh, rename to short names, and track originals for updates.

Run repository linting and formatting checks, fix lint failures, and verify a clean result. Use when asked to lint code, resolve style/tooling violations, or make CI lint checks pass across changed files or whole repositories.

Evaluate and improve performance by identifying hot paths, reducing unnecessary allocations/work, and optimizing data flow and I/O behavior without changing required functionality. Use when addressing latency, throughput, CPU/memory overhead, startup cost, or scaling bottlenecks.

Create, maintain, and execute ExecPlans for complex or multi-step work that spans many files or repositories. Use when work needs a self-contained living plan with milestones, decision tracking, concrete commands, validation criteria, and restartable context.

Create and manage pull requests with submodule-aware branch handling, auto-generated titles and descriptions from commit history, and safety-first push behavior. Use when asked to open a PR, check PR status, push a branch, or prepare changes for review.

Refactor and improve existing code while preserving behavior, using incremental, verifiable changes to reduce complexity and improve maintainability. Use when updating existing implementations, simplifying architecture, or restructuring legacy code without changing external contracts unless explicitly requested.

Assess and improve runtime reliability by hardening error handling, resource lifecycle management, cancellation/timeout behavior, and recovery characteristics. Use when building or reviewing code that touches I/O, concurrency, retries, long-running jobs, cleanup paths, or failure handling.

Safely rename symbols and identifiers with scope control, ambiguity checks, and post-rename verification. Use when renaming variables, functions, types, files, or API fields across one or many files, especially when consistency and low false-positive risk matter.

Create and update research documents in research/ for technical investigations, vendor evaluations, architecture decisions, and product research. Use when asked to start research, write up findings, or update an existing research doc.

Perform structured code review focused on correctness, test coverage, and regressions, with findings prioritized by severity and grounded in file/line evidence. Use when asked to review, audit, assess risk, or evaluate change quality before merge or commit.

Assess and improve security posture of code changes by addressing input handling, authentication/authorization boundaries, secret management, cryptography usage, and dependency/supply-chain risk. Use when code processes untrusted input, handles credentials/tokens, accesses protected resources, or changes external interfaces.

Sync local repo after a PR merge by switching to main, fetching with prune, pulling, and deleting stale local branches. Use when asked to sync, clean up after merge, or delete merged branches.

Discover, document, resolve, and maintain tech debt items in per-repo TECH_DEBT.md files. Use when asked to find tech debt, document debt, fix a tech debt item, or audit a submodule for debt.

Design and implement high-value tests for changed or specified behavior, including edge cases, error paths, and dependency interactions, with a plan-first workflow and explicit verification results. Use when asked to add tests, improve coverage, validate fixes, or harden behavior.