dots

Draft or update high-quality technical documentation for the Obsidian vault with clear structure, Obsidian-ready frontmatter, appropriate tags, rationale, and practical examples.

End-of-session documentation: synthesizes what was done, why, patterns implemented, and codebase knowledge gained, then hands off to chronicle-docs to write a structured Obsidian note.

Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria.

Review a colleague's pull request and output a structured review with actionable findings — no auto-commenting on GitHub, no plan mode, no implementation. TRIGGER when: user asks to review a PR, review a pull request, check a PR, look at a PR, give feedback on a PR, first-pass review, or references a PR number/URL to review. DO NOT TRIGGER when: user asks to fetch or summarize existing PR comments (use get-pr-comments), or wants to review their own uncommitted changes (use review-local).

Create a pull request OR update the description of an existing PR for the current branch. Handles repo PR templates, JIRA ticket extraction (including title prefix), a prose summary followed by implementation bullets with rationale, testing evidence, and screenshot prompts for UI/frontend changes. TRIGGER when: user asks to create a PR, open a PR, "/describe-pr", "/create-pr", update the PR description, refresh PR body, or after pushing a branch that already has a PR.

Fetch review and discussion comments for the current pull request, classify human-reviewer feedback against tracked patterns, and turn it all into a concise action list.

Add or remove CLI tool patterns from Claude Code's sandbox permissions allowlist and denylist. Triggered proactively by user request or reactively when a Bash command is blocked.

Review the latest Claude Code documentation, compare it against the current Claude Code configuration, and recommend specific improvements based on new or underused features. Run periodically (e.g. monthly) to catch new capabilities that could improve the workflow. TRIGGER when: user asks to audit, review, or check their Claude Code setup / config / skills / hooks / MCP servers / settings, or asks "what new Claude Code features should I be using". DO NOT TRIGGER for: Codex setup audits, app-code audits, or PR/local-change reviews (use review-pr or review-local).

Review user feedback or session patterns and create or update agent skills so durable behavior changes are captured in the right SKILL.md with proposal-first approval.

Clean up a patch before review or commit by removing AI-style slop, checking test value, validating key paths, and surfacing remaining risks.

Review the user's own unmerged work for bugs, regressions, risk, and missing tests with severity-first output and concrete file references. Handles two scopes: uncommitted/staged changes (pre-commit) and the full branch diff against its base (pre-PR self-review). Read-only — produces findings, does not edit code. TRIGGER when: user asks to review their local changes, uncommitted changes, staged changes, current diff, current branch, "review my branch", "review my changes", or "review before I open the PR". DO NOT TRIGGER when: user references an existing PR number/URL (use review-pr), wants a security-only audit (use built-in security-review), or wants the patch tidied/edited rather than reviewed (use pre-commit-cleanup).

Step-by-step tutorial guidance: the AI explains, the user learns. Use when the user asks to be taught, walked through an implementation, or helped to learn a new language, codebase, or stack — either before writing the change (guide them through it) or after the fact (explain what was done so they can review and understand it). Audience is a TypeScript-comfortable engineer learning something new. Do not edit files or write the implementation.