idle-skills

Plugin idle-skills from marketplace idle-agent-tools in ambrovia/idle-agent-tools. Commands read from the repo's manifest — 24 days ago.

Skills (15)

• architecture-critique

  Read-only critique of a proposed technical breakdown against the plan, before it is accepted. Checks alignment, necessary contracts, feasibility, simplicity and readability. Requested when worth challenging, never scheduled; reports blockers without scoring or rewriting.

• architecture

  Work out the technical shape of an agreed plan — contracts, types, schemas, dependency order — and propose it as tasks and decisions. Use only where scope and complexity make those definitions necessary. Never decides how the program works; that is the plan's job.

• design

  Decide consequential UX/UI behavior for an agreed task. Use when a changed user-facing surface is not already determined by an existing approved pattern and a design system is configured. Skip backend/infra and trivial pattern-following changes.

• lore

  Capture, scan, or index terse, currently actionable tribal knowledge about non-obvious cross-cutting constraints, workarounds, and gotchas. Use when missing context would cause a future mistake; never use lore as history, a changelog, or a second requirements system.

• orchestrate

  Drive one or more tasks to a CI-green PR, through whatever shape each one actually needs. Runs the interviews itself, breaks the work into goals with the maintainer, and dispatches the rest. Never creates scope.

• program-design

  Interview the maintainer about how the program should work, and write it into the plan in plain words. Use when the approach is not obvious. Produces the how-it-works half of the plan — no contracts, types, or schemas.

• refine

  Interview the maintainer about what they need from a task, and write it into the task in their words. Use when a dropped task's value, boundary, or a load-bearing noun is not already settled. Skip when it is already clear enough to build from.

• retro

  Record concrete successes and friction after a task, session, or pipeline run. Observe from artifacts and trajectories without fixing, assigning work, or changing process.

• review

  Read-only review of implemented work against its task's goal or an explicit changed-file scope. Judges whether the goal is reached, applicable constraints/rules, regressions, and plausible changed risks. Produces operational findings and a verdict; never edits code.

• self-improve

  Review accumulated retros, outcomes, interviews, and decisions to identify recurring evidence-backed patterns and propose surgical rule, memory, or skill changes for human approval. Works both when triggered manually and on a weekly schedule. Never applies changes automatically.

• setup

  Configure evidence-backed, maintainer-approved pipeline rule slots and design-system paths for this repository. Use during installation or when repository conventions change; do not promote generic best practices into binding policy automatically.

• ship

  Turn completed work into a clean CI-green merge-ready PR. Verifies, commits, updates the PR, and waits for CI; does not merge. Review belongs to the orchestrate skill.

• write-code

  Implement a claimed task with the smallest clear solution that reaches its goal. Use after the plan is agreed and the task is yours.

• write-docs

  Write or review user-facing documentation when it is an explicit deliverable or when an implementation makes authoritative documentation false. Preserve project voice and factual accuracy; use reading-funnel craft proportionately to page type.

• write-tests

  Create proportionate failing evidence for a task's goal before implementation when behavior is regressible and an automated test is appropriate. Use in the build phase; do not let tests create requirements.

Subagents (4)

• README
• idle-builder

  Implementation producer. Use to write tests and production code for a claimed task, or to fix what a task came back for. Executes approved outcomes and contracts without redesigning scope.

• idle-planner

  The planner: holds the whole picture of a task tree and plans every level of it — goals, scope, order, UI/UX decisions, technical shape. Use when work needs structuring or re-planning; do not use for formal critique, implementation, or scope creation.

• idle-reviewer

  Independent read-only evaluator for requirements, design, architecture, and implemented code. Use when a critique or review requests evaluation. Produces evidence-backed findings only; never authors or repairs the evaluated work.

Hooks (4)

• SessionStart

  bash -lc 'for root in "${CLAUDE_PLUGIN_ROOT:-}" "${CODEX_PLUGIN_ROOT:-}" "$HOME"/.claude/plugins/cache/idle-agent-tools/idle-skills/* "$HOME"/.codex/plugins/cache/idle-agent-tools/idle-skills/* "$HOME

• PostToolUse

  bash -lc 'for root in "${CLAUDE_PLUGIN_ROOT:-}" "${CODEX_PLUGIN_ROOT:-}" "$HOME"/.claude/plugins/cache/idle-agent-tools/idle-skills/* "$HOME"/.codex/plugins/cache/idle-agent-tools/idle-skills/* "$HOME

• PostToolUse

  bash -lc 'for root in "${CLAUDE_PLUGIN_ROOT:-}" "${CODEX_PLUGIN_ROOT:-}" "$HOME"/.claude/plugins/cache/idle-agent-tools/idle-skills/* "$HOME"/.codex/plugins/cache/idle-agent-tools/idle-skills/* "$HOME

• SubagentStart

  bash -lc 'for root in "${CLAUDE_PLUGIN_ROOT:-}" "${CODEX_PLUGIN_ROOT:-}" "$HOME"/.claude/plugins/cache/idle-agent-tools/idle-skills/* "$HOME"/.codex/plugins/cache/idle-agent-tools/idle-skills/* "$HOME

Similar plugins