conn-castle

Run completed local work through PR delivery: audit changes, verify locally, commit, push, open PR, monitor CI, handle review comments, finish green, then merge after approval and clean up the branch. Use `fix-ci` for failing PR checks.

Audit the existing test suite for redundancy, quality gaps, and organization: duplicate or self-confirming tests, coverage gaps metrics miss, and safe cleanup fixes. Use `prune-new-tests` instead when the target is tests added in the current uncommitted diff.

Raise coverage: find the real coverage command, select under-covered files, add behavior-focused tests, iterate to target or blocker. Use `audit-tests` for full-suite cleanup; use `prune-new-tests` to prune speculative tests in the current uncommitted diff.

Use `al dispatch` for focused headless second-agent work — another model, tool set, skill, or review perspective. Trigger when the user asks to run al dispatch, use a subagent or second agent, get a fresh-context review, or invoke a target agent with a skill. Do not use for ordinary shell commands, local test runs, web search, browser automation, or multi-agent orchestration.

Use `npx ctx7` to fetch current documentation, API references, and code examples for libraries, frameworks, and APIs when the answer is version-dependent or post-cutoff, or local sources (README, docs/, CLI help) can't answer it. Trigger for API syntax, configuration, setup, migration, or "how do I use <technology>" questions. Do not use for generic web research, local file reading, browser automation, or package installation.

Use playwright-cli for real-browser UI automation, screenshots, Playwright test debugging, and interactive test generation. Trigger when a task needs browser interaction, UI inspection, end-to-end verification, or Playwright test repair. Do not use for generic test running, web search, non-browser docs, or API-only work.

Use the Tavily CLI for web search, URL extraction, site URL discovery, and cited research. Trigger when the user needs current web information, asks to search or read a webpage, provides URLs to inspect, or wants a sourced research report. Do not use for browser automation, local docs, or Tavily setup.

Burden-of-proof review of tests added in the current uncommitted diff: auto-delete any new test that can't justify itself with a production-code mutation that would flip its assertion. Use `audit-tests` for full-suite health; use `boost-coverage` to add tests.

Audit and iteratively fix uncommitted working-tree changes until a confirming review finds no remaining actionable issues. Use to stabilize in-progress diffs and report each review/fix round with severity.

Audit the existing test suite for redundancy, quality gaps, and organization across unit, integration, and e2e tiers. Finds duplicate or self-confirming tests, coverage gaps metrics miss, and safe cleanup fixes. Use `prune-new-tests` instead when the target is tests added in the current uncommitted diff.

Raise coverage by finding the real coverage command, selecting under-covered files, adding behavior-focused tests, and iterating to target or blocker. Use `audit-tests` for full-suite cleanup; use `prune-new-tests` to prune speculative tests in the current uncommitted diff.

Execute an existing plan/task/context artifact set: make aligned code changes and verify code, tests, docs, and memory before closeout. Use `plan-work` when no plan exists.

Write a scoped implementation plan, task list, and context file for a requested change, roadmap slice, execution strategy, task breakdown, or pre-implementation design artifact.

Burden-of-proof review of tests added in the current uncommitted diff: auto-delete any newly added test that cannot defend its existence with a concrete production-code mutation that would flip its assertion. Use `audit-tests` for full-suite health; use `boost-coverage` to add tests.

Run completed local work through PR delivery: audit changes, verify locally, commit, push, open PR, monitor CI, handle review comments, and finish green. After exact approval (`I approve merging PR #<N>`), merge the PR and clean up its source branch. Use `fix-ci` for failing PR checks.

Codebase-wide complexity reduction: remove dead code, simplify overly complex functions, and split files with mixed responsibilities across explicit paths or the whole repo. Use `simplify-new-code` instead when the target is the current uncommitted diff.