do-it

Plugin do-it from marketplace fws-skills in flyingwebie/skills. Commands read from the repo's manifest — 1 month ago.

Skills (13)

• build-loop

  Execute Do-it tasks. Reads .do-it/specs/ (current truth) and .do-it/changes/ (proposals), picks the next buildable band, builds under TDD discipline (software) or acceptance-driven (generic), respects the UI-first layer gate, and keeps the loop memo current. Product review runs per feature; optional

• design

  Emit the compact design state the build loop consumes: tokens, the F0 foundation manifest, and per-entity contracts. Narrative design docs defer to --full.

• do-it

  Run the Do-it spec-driven MVP workflow in Codex. Use for PRD, brownfield mapping, config migration, grilling, design, preview, building, review, iteration, project status, retrospectives, or autonomous delivery.

• grill

  Adversarially challenge a spec-layer artifact — PRD, change proposal, or design system — before code exists. Hunts gaps, contradictions, ambiguity, scope creep. Triggered by /do-it:grill or "challenge the PRD", "poke holes".

• iterate-fix

  Apply fixes after a REVISE verdict, or after a task's mechanical gate goes red. Root-cause discipline only — no symptom fixes, no test-softening. Triggered by /iterate, or by the Ralph loop on a red gate or a REVISE.

• llm-wiki

  Maintain docs/ as an optional LLM Wiki. Direct Pass/Query always work; automatic passes run only when project knowledge maintenance enables them.

• loop-runner

  Set up and explain Do-it's autonomous drivers — /do-it:loop (Claude Code's native loop, deterministic stop via scripts/loop-guard.sh) and /do-it:ralph (the legacy in-session Stop hook that re-feeds the iteration contract). Preflight, settings resolution, the stop report, and stop semantics for both.

• map-codebase

  Walk an existing codebase with parallel mapper subagents into .do-it/codebase/ — 7 files: stack, integrations, architecture, structure, conventions, testing, concerns. Detects and migrates GSD, Superpowers, or Ralph Loop v0.1 state; optional wiki seeding follows knowledge maintenance. Run by /do-it:

• prd-generator

  Use when the user wants to start a new project from a PRD. Triggers include "create a PRD", "write product requirements", "build a product spec", "clone a product", "replicate a product", "generate a PRD for", or any /prd command in the Do-it plugin. Produces .do-it/specs/prd.md as the foundation do

• retrospect

  Mine progress, metrics, review findings and failed attempts into durable lessons. Direct runs always work; automatic archive runs require knowledge maintenance to be enabled.

• review-work

  Adversarial review of a completed FEATURE (default) or a single task — spec compliance and quality+security as two fresh subagents in parallel, dispatched with the briefs in references/briefs/. Used by /do-it:review, and by build-loop before a change is archived.

• status

  Read-only briefing of the current Do-it project — features, current change, gates, live preview, spec coverage, recent activity, suggested next move. Run by /do-it:status. Never writes anything.

• worktrees

  Run each open Do-it change in an isolated git worktree under .worktrees/. Substrate for parallel features — required whenever the loop fans out (subagent_concurrency > 1), also usable serially via _meta.json use_worktrees.

Slash commands (16)

• /do-it:build

  Build the next band of tasks. TDD for software, acceptance-driven for generic, with product review per feature and optional knowledge maintenance.

• /do-it:cancel-ralph

  Cancel any active Do-it autonomous loop — the native /do-it:loop run or the legacy Ralph Stop-hook loop.

• /do-it:config

  Review and safely upgrade .do-it/specs/_meta.json to the latest Do-it schema and flags.

• /do-it:design

  Emit tokens, the F0 foundation manifest, and entity contracts so the UI base can render before F1.

• /do-it:do

  Do-it entry point. Routes to the right next step based on current project state.

• /do-it:grill

  Adversarially challenge whatever's in front — PRD, change proposal, or design spec. Hunts gaps, contradictions, scope creep before code is written.

• /do-it:iterate

  Apply review fixes. Root-cause only — no symptom fixes.

• /do-it:loop

  Run Do-it autonomously until the MVP is done. Native loop, deterministic stop condition, parallel subagent bands.

• /do-it:map

  Map an existing codebase with parallel mapper subagents. Migrates GSD/Superpowers/Ralph state if found.

• /do-it:prd

  Generate a PRD (interactive). The PRD is the foundation of the whole loop.

• /do-it:preview

  Start (or reuse) a warm preview of the app and screenshot it. The fastest way to see what has been built.

• /do-it:ralph

  Start the Stop-hook autonomous loop (legacy driver — prefer /do-it:loop). Builds band by band until done, a gate, or max iterations.

• /do-it:retrospect

  Mine the run's evidence into AGENTS.md + lessons — the loop's ITERATE stage

• /do-it:review

  Adversarial review of a completed feature (or one task) — spec compliance and quality+security, in parallel.

• /do-it:status

  Read-only project status briefing. Shows features, current change, gates, spec coverage, and the next suggested command.

• /do-it:statusline

  Install (or remove) the Do-it live status line into your Claude Code settings. Required because plugins can't ship a top-level statusLine.

Subagents (3)

• builder-cheap

  Build one mechanically-verified Do-it task (ui-types, ui-fixtures) on the cheapest tier. Admitted only where a script catches a skipped test.

• builder-deep

  Build one Do-it task on the highest tier. Opt-in only, via _meta.json model_tiers — never a shipped default.

• builder

  Build exactly one Do-it task under the brief it is given. Default builder tier — inherits the session model.

Hooks (2)

• Stop

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/stop-hook.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/pre-commit-branch-guard.sh"

Similar plugins