agent-skills

Rewrites robotic or AI-sounding text to feel natural while preserving meaning and voice. Use when humanizing or reviewing generated, over-polished, repetitive, or impersonal prose. Covers 24 wording patterns and 14 construction patterns.

Writes, rewrites, drafts, or reviews text in Lachlan's written and spoken voice. Use when the user asks to "talk like me" or "sound like me", or needs Slack replies, emails, leadership notes, technical feedback, customer or strategy messages, meeting comments, or talking points that match Lachlan's directness, dry humour, and evidence-first engineering judgement.

Iteratively review and improve the current PR or branch by running general-code-reviewing, fixing grounded issues, validating, and re-reviewing until the change is merge-ready or a blocker remains. Use when asked to auto-review, self-review and fix, harden, polish, get a PR ready, or iterate on code quality rather than only return review comments.

Draft and revise durable engineering plans. Use for `docs/plans`, architecture, rollout, design docs, first-slice selection, open-question resolution, adversarial pressure-testing, and plan maintenance during implementation.

Run the full code-to-merged-PR workflow: auto-review the current branch or PR, create or update a GitHub PR, babysit it through reviews and CI, merge it, and report the final state. Use when asked to land a PR, land this, ship this branch, get this into a PR and merge it, or otherwise carry code from local changes or a current branch to a merged PR.

Read x.com/twitter.com posts, tweets, quote tweets, and user-provided post URLs through the authenticated xurl CLI, with Birdclaw as the local cache and research fallback. Use when asked to read, summarize, inspect, quote, fetch, or explain an X/Twitter post URL or status ID; when the user says to use their Twitter account; or when read-only Twitter/X context should come from xurl or Birdclaw instead of web search.

Operate Birdclaw, the local-first Twitter/X memory CLI and web app backed by SQLite, plus xurl for direct X API reads. Use when asked to read x.com/twitter.com post URLs or post IDs, set up or troubleshoot Birdclaw, inspect its local store, sync or read bookmarks/likes/mentions/timelines/DMs, use xurl or bird transports, import a Twitter archive, start the local Birdclaw UI, or build bookmark research briefs.

Write, review, and optimize high-performance Zig codebases with measurement-first workflow, Ghostty and Mitchell Hashimoto-inspired patterns, cache-aware struct layout, allocator and pool design, comptime specialization, SIMD/scalar fast paths, C ABI boundaries, and regression checks. Use when Codex is working on `.zig` changes where throughput, latency, startup time, memory use, cache behavior, parser/rendering loops, atomics, allocators, binary size, or layout-sensitive code matter.

Execute engineering plans locally, slice by slice. Use when asked to implement, continue, or iterate on an existing plan by reading/updating it, working in a local branch, validating, running auto-review, committing slices, and only publishing or babysitting a PR when requested.

Produce a concrete end-of-work walkthrough after implementation, plan execution, PR preparation, or a completed local branch slice. Use when the user asks for a walkthrough, demo, handoff, summary of completed work, what changed and why, UX or CLI examples, local web showcase, impact, unexpected difficulty, or next suggested steps; also use as the final handoff step from executing-plans.

Check whether repository docs, plans, examples, runbooks, changelogs, generated docs, or other durable written artifacts are updated for a code or configuration change. Use when asked to verify docs or plans are current, prevent documentation drift before review or merge, update docs after implementation, or check whether a PR needs documentation changes.

Check whether a pull request title and body still match the actual diff, current scope, reviewer-relevant impact, and repository PR-description conventions. Use when asked to check, update, rewrite, or refresh a PR description or title; verify PR metadata before review or merge; or fix stale PR text after commits changed the branch.

Babysits GitHub pull requests through review comments, Codex feedback, rebases, Buildkite failures, merge queues, and final merge. Use when asked to babysit a PR, get a PR merged, fix PR feedback, or keep working until a PR is mergeable.

Handles Codex GitHub PR review loops by waiting for reviews, fixing actionable feedback, resolving threads, and requiring Codex's main-thread thumbs-up. Use when Codex is a reviewer, a PR has `@codex review` activity, or Codex has added an `eyes` reaction to the PR description or latest review trigger.

Orchestrate a broad code review by running ship-risk and maintainability review passes, then synthesizing both. Supports sub-agent delegation only when the user explicitly asks for sub-agents, delegation, or parallel review work. Use for general PR reviews, diff reviews, code reviews, or broad review requests when the user did not ask for only one narrower review lens.

Run an extremely strict maintainability review for abstraction quality, giant files, and spaghetti-condition growth. Use for a thermo-nuclear code quality review, thermonuclear review, deep code quality audit, or especially harsh maintainability review.

Find deepening opportunities in a codebase, informed by engineering plans in docs/plans/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.

Uses an Amp-style Librarian workflow to explain repositories outside the current workspace. Use when tracing dependency internals, comparing implementations across GitHub repositories, reading remote commit history, or understanding external architecture from a non-Amp agent.

Performs skeptical, high-signal code reviews that try to break confidence in a change by surfacing grounded, material failure modes. Use when asked for an adversarial review, ship/no-ship assessment, or a review focused on subtle production risks instead of balanced feedback.

Build and review Go command-line applications with idiomatic package layout, Kong command parsing, charmbracelet/log logging, config loading, stdout and stderr behavior, linting, tool version management, and behavior-first tests. Use when Codex is creating or refactoring `main.go`, `cmd/` command trees, global flags, subcommands, help text, config handling, or CLI UX in Go.

Write, review, and refactor Go code with idiomatic package design, naming, error handling, concurrency, performance, testing, linting, and toolchain management. Use when Codex is implementing or reviewing `.go` changes, adding Go tests or benchmarks, setting up lint or version tooling, reshaping package APIs, choosing between Go patterns, or applying current version-gated Go guidance while respecting the repository's actual toolchain and conventions.

Responds to PR review comments. Use when addressing reviewer feedback, replying to diff comments, or checking for unaddressed reviews.

Search and manage Linear issues. Use for listing issues, creating/updating issues, changing status, viewing issue details, or any Linear workflow.

Manage Notion pages, databases, and comments from the command line. Search, view, create, and edit content in your Notion workspace.

Read Slack messages, threads, and channels via CLI. Use when asked to view Slack URLs, search Slack, or look up Slack users.