Comprehensive multi-phase PR review — change analysis with risk_class routing, 2-pass deep code review, risk-gated quality-review passes (domain-architecture, data-contracts, ops-reliability, change-evolution), security scanning (OWASP deepen), and documentation-drift checks, posted as one unified GitHub review. Test execution and coverage analysis are left to CI. Plus fix: apply the findings in the working tree — refactor each code smell by the named Fowler/Kerievsky move in the language's own idiom, rewrite flagged docs and comments, regenerate stale artifacts, and defer anything needing a human design decision, one commit per finding and the test suite once per partition, then push the head branch once it is green and reply to and resolve the review comments it answered. Plus monitor-pr: continuously watch a PR for CI failures and review comments (drive with /loop), delegate the fixing to fix, push, and resolve or reply-acknowledge each comment once its fix ships. Rust and mac-ui Web Components changes get tailored checks.
11 skills
Human-facing developer writing, done well, as a pipeline: write-doc or writing-design-docs drafts, review-doc scores against the Practical Prose rubric, fix-doc applies the findings and strips AI tells. Docs that describe structure or flow get a diagram (architecture-design if installed, else Mermaid). create-slides builds a minimalist Machinify HTML deck from a doc or report: one claim and one visual per slide, prose in speaker notes, every diagram drawn by architecture-design's create-diagram. Ships the laconic output style: write the conclusion, not the path to it. Adapted from ceej-skills, Practical Prose (jlevy/practical-prose), humanizer (blader/humanizer), pertinent, and a private slide-deck template, under MIT and CC BY 4.0.
5 skills
Relentless interviews to stress-test a plan, decision, or design before you commit to it. Provides grilling (the core interview), grill-me (a quick trigger), grill-with-docs (grilling that also produces ADRs and a glossary as decisions crystallize), and wayfinder (plan work too large for one session as a map of decision tickets, resolved one at a time).
4 skills
Query Claude Code session history and conversation data with SQL (DuckDB) via cc-query. Provides the reflect skill for session analysis, token usage, tool patterns, and agent activity, and plugin-curation for evidence-based keep/consolidate/prune audits of installed plugins, skills, and MCP servers.
2 skills · hooks
Manage the full lifecycle of a Claude Code working session. start-session opens a task in a fresh, isolated git worktree (via EnterWorktree, never `git worktree add`) named after the argument on branch `worktree-<name>`, reports what untracked state the worktree inherited — `.worktreeinclude` copies, symlinked node_modules — and whether its `.env` still points at the main checkout's stack, and loads the JIRA story when the name is an issue key. exit-session tears down the ephemeral resources a session created — background processes and dev servers, Docker/compose stacks with their volumes, networks and locally built images, local Kubernetes, git worktrees (the session's own and every one it handed to a subagent, unlocking them first), throwaway local branches, and temp files — permission-free, then quits the CLI. Worktrees are attributed by the pid in their git lock reason and Docker resources by their compose-project label, so teardown survives a compacted context. Runs prompt-free in default, accept-edits and auto mode; a bundled UserPromptSubmit hook refuses only plan mode, where mutating tools are off. It hands back the deny-listed remote/history ops rather than running them.
2 skills · hooks
Fabric skins for Claude Code: tool rows that open onto their input and answer, diff, table, code, math, Mermaid and shell cards and a calm mode, in the Fabric palette (one chroma, ink on near-black or paper). Ships the fabric skin plus sixteen upstream themes; /skin swaps them live. A fork of hellosverre/claude-skins; the same tokens draw architecture-design's diagrams.
1 skill