downshift

Plugin downshift from marketplace downshift in tiagovilasboas/downshift-ai. Commands read from the repo's manifest — today.

Skills (10)

• downshift-layering

  Use before adding or changing a Downshift feature, adapter, hook, command or policy. Enforces the layering rule so no change serves only one harness.

• architecture-coverage

  Fill the architecture coverage inventory before proposing a roadmap phase that touches the Downshift binary. Triggers on: roadmap, architecture, BDI, accounting, compaction, "the whole thing", or any plan with multiple phases. Only applies when the workspace is the Downshift repository.

• downshift-audit

  Read-only audit of a Downshift install and its fixed agent context. Use when the user names Downshift and asks why it does not rewrite models, whether its hooks are wired, or where subagent tokens go; the fixed-context measurement is part of that audit, not a standalone trigger.

• downshift-effort

  Advice for the orchestrator on matching reasoning and verification effort to what a task puts at risk. Use when deciding how hard to think, how much to check, or what effort to give a subagent, with Downshift installed.

• downshift-orchestrator

  Playbook for the orchestrator (the lead session) when it delegates work to subagents. Use when splitting a task into subagents or choosing the model and effort of a spawn, with Downshift installed. Also use when the user says "downshift full power" or "full-power", "potência total" or "potencia-tota

• downshift-setup

  Install or repair Downshift hooks on the user's machine for any harness (Claude Code, Cursor, Codex, Antigravity, KiroCrew). Use when asked to install, set up, wire, fix or undo Downshift, or when subagents never change model. Dry-run first, write only after a yes.

• downshift-squad

  Invoke Downshift agent squads for coordinated multi-agent work. Use when the user says "squad", "time", "convoca", or wants to delegate to specialized agents. For "downshift full power", "força total" or "downshift shift <squad>", use the downshift-orchestrator skill.

• downshift-tune

  Tune what a session spends in tokens and quota with Downshift installed, in a fixed order with evidence first. Diagnoses routing, fixed context, tool output and project policy, then proposes one change at a time. Use only when the user names Downshift, or asks to cut subagent tokens, context or quot

• harness-engineer

  Plan and split multi-step agent work with Downshift. Use when the user names Downshift and asks how to split or parallelize subagent tasks, or why delegated work is not getting a smaller model.

• ponytail

  Lazy senior dev mode: the smallest change that fully solves the task, and a reply a busy human understands in one read. Use on any coding task (writing, fixing, refactoring, reviewing, choosing dependencies) and when the user says "ponytail", "be lazy", "simplest solution", "yagni", or complains abo

Slash commands (1)

• /downshift:status

  Session status: subagent tree, routing reasons, pause and resume

Subagents (15)

• ai

  AI/LLM architecture, model routing, agent orchestration, and FinOps

• architect

  System design, cross-layer architecture, trade-offs, and interfaces

• coder-back

  Backend specialist: writes clean, tested server-side code in Go/Node/PHP, API contracts, SQLite/MySQL persistence, concurrency, idempotency, and edge security.

• coder-front

  Frontend and visual swarm specialist: React 19, Next.js, Tailwind 4, Canvas 2D, 60fps Web Workers, Swapo-style agent swarm visual effects (cursors, neon lasers, Tour Pill, particles), offline PWA, accessibility, and bundle budget.

• coder

  Full-Stack Integrator: vertical slices, end-to-end integration, and stitching backend APIs with frontend UI components. TDD when appropriate, minimal diffs.

• debugger

  Debug and E2E test specialist: investigates test failures, flaky tests, E2E issues, and runtime bugs. Expert in test frameworks, browser automation, and failure analysis.

• documenter

  Technical writing, README, API docs, guides, and editorial review

• mechanical-worker

  Bounded mechanical work in this repository: a rename, a typo fix, a small edit in one file, running one test command and reporting the result. Use for well-specified, low-risk tasks so the parent keeps its context clean.

• performance

  Performance analysis specialist: profiles code, identifies bottlenecks, analyzes complexity, suggests optimizations. Read-only analysis.

• platform

  Backend infrastructure, DevOps, CI/CD, databases, and reliability

• refactor

  Refactoring specialist: improves code structure without changing behavior. Renames, extracts, simplifies, removes duplication. Preserves tests.

• researcher

  Code analysis and research specialist: understands codebases, finds patterns, traces dependencies, answers architectural questions. Read-only exploration.

• reviewer

  Code review, test strategy, quality gates, and verification evidence

• security

  Security audit, threat modeling, vulnerability detection, and trust boundaries

• tester

  Test runner and coverage analyst: runs test suites, reports coverage, identifies gaps, suggests test cases. Does not write tests (that's coder's job).

Hooks (8)

• PreToolUse

  downshift claude-code

• PreToolUse

  downshift claude-code

• PostToolUse

  downshift claude-code-post-tool-use

• PostToolUse

  downshift claude-code-post-tool-use

• PostToolUse

  downshift claude-code-post-tool-use

• PostToolUseFailure

  downshift claude-code-post-tool-use

• SubagentStop

  downshift claude-code-subagent-stop

• SessionStart

  downshift claude-code-session-start

Similar plugins