agentic-toolkit

Use when looking for the next ticket to work on. Detects the project's ticket system (GitHub Issues, Jira, GitLab Issues, Azure Boards, etc.), fetches actionable open tickets that are unassigned or assigned to you, scores by severity/simplicity/value/blocking-power/dependencies, picks the best candidate, claims it by self-assigning (team-safe), branches, implements, tests, formats, and waits for review with a UI testing tip. Accepts an optional ticket ID argument (e.g., `/next-ticket 42`, `/next-ticket PROJ-42`) to skip evaluation and pick up a specific ticket directly.

Format, lint, test, commit, push, and create a pull request. The single "I'm done" command.

Compress markdown files into concise prose to save input tokens. Default mode reduces verbosity without losing information. With deep, verifies each section against the codebase first, removing stale or incorrect content before compressing. Trigger: /compress-markdown [deep] <filepath>

Craft a well-researched, well-structured ticket from a user-provided idea. Explores project context when available, researches prior art, asks clarifying questions only when options are too nuanced to auto-resolve, deduplicates against the existing backlog, and files one world-class ticket. Works for new or mature codebases.

Read PR review comments, validate against code, fix valid ones, push, resolve addressed threads, and explain unresolved ones.

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

Convert a git worktree into a local branch. Rebase onto latest default branch, clean up project resources, remove worktree, checkout branch in main workspace. Use when finishing work in a worktree.

Commit, push, create/merge PR, sync local default branch, delete branch. The standard "I'm done with this branch" workflow.

Updates project dependencies. Checks open bot PRs for CVE patches, applies safe minor/patch updates, and researches/fixes breaking changes for major bumps. Optional scope (frontend|backend|infra|all) and major flag.

Use when work on a branch is complete or in-progress and you want to re-evaluate the approach from scratch. Performs retrospective analysis, re-architects to reduce complexity and fragmentation, auto-implements without committing, and outputs a brief testing playbook for validation.

Audits a codebase for bugs, security vulnerabilities, missing error handling, race conditions, architectural gaps, DRY violations, naming inconsistencies, incomplete implementations, and robustness issues. Caches tickets to disk, then spawns 4 parallel sub-agents (one per focus cluster) to scrutinize and file tickets.

Investigates a codebase for proven defects using adversarial 4-pass analysis (frame, trace, falsify, prove). Caches tickets to disk, then spawns 4 parallel sub-agents (one per bug category) to find, prove, and document bugs with enough rigor that a skeptical maintainer could fix each from the report alone.

Audits a project for UX gaps, broken workflows, missing states, confusing terminology, visual inconsistency, navigation/state issues, destructive action safety, data presentation, accessibility issues, and competitive table stakes. Caches tickets to disk, then spawns 4 parallel sub-agents (one per focus cluster) to scrutinize and file tickets.