spec-workflow

Plugin spec-workflow from marketplace development-skills in Zugruul/development-skills. Commands read from the repo's manifest — 3 days ago.

Skills (37)

• agent-identities

  Shows or configures which git author identities (name/email) the agent roles — dev, reviewer, orchestrator, peer-reviewer — commit as. Templates resolve per clone ({name}, {local}+suffix@{domain} placeholders fill in from your own git config), so commits stay attributable without hardcoding anyone's

• ask-brain

  Asks a repo's identity brains collectively — every role that has a brain directory under .neural-network/identities/, not just one — a question, and answers it grounded in what those roles have actually learned, without running a build iteration or touching the board. Use for '/spec-workflow:ask-bra

• ask-identity

  Asks one identity's brain — dev, reviewer, orchestrator, or any custom role with a brain directory under .neural-network/identities/ — a question, and answers it grounded in what that one role has learned, without running a build iteration or touching the board. Use for '/spec-workflow:ask-identity'

• auto-merge

  Checks the status of, turns on, or turns off auto-merge mode, where an agent reviews, approves, and merges pull requests instead of a human doing it. Use with 'status' (default), 'on', or 'off' — when the user asks whether auto-merge is active, wants the loop to merge without them, or wants human ap

• board

  Reads and updates the GitHub Project board configured in .neural-network/project.yaml — pick the next task, show an issue with its human comments, move it between statuses, set priority or estimate, reply to comments, or file a bug, all from this session. Use when the user mentions the board, task s

• brain

  Inspects and tends the per-identity zettel brains — status, recall, mint, prune, directory, graduate-check, explain, path — all via brain.sh. This is orchestrator-only memory (each role's brain is private; subagents never read one). Use to see what a role has learned, recall lessons for a task, mint

• build-next

  Runs ONE autonomous build-loop iteration end to end — checks the checkpoint, picks the next board task (reading its human comments first), implements it via TDD to a recorded green gate, opens a PR, and keeps the board current throughout. Use for 'build next', 'continue/resume the loop', 'work the b

• changelog-generate

  Regenerates the repo's CHANGELOG.md from git history, organized by release version and grouped by change type (features, fixes, etc.). Use when asked to generate, update, or refresh the changelog, to see what changed in a given release, or to summarize recent commits into release notes.

• checkpoint

  Pauses or resumes the autonomous build loop by writing or clearing a local gitignored flag file, always at a safe task boundary and with a handoff written first. Use for 'pause the loop', 'stop after this task', 'resume the loop', or holding work before a review/deploy gate.

• comfyui-render

  Queue a pinned ComfyUI image workflow on an explicitly provisioned SSH target, inspect task progress, or request safe cancellation of an owned assistant render. Uses the assistant task engine; excludes human remote-compute CLI jobs and workflow authoring.

• concurrency

  Shows or sets how many tasks the build loop works on at once (methodology.maxInProgress) — this one number is both the board's WIP limit and the number of parallel implementation lanes. Use when the user asks how many tasks run at once, wants parallel lanes, or wants strictly-sequential building bac

• craft-spec

  Helps you write a spec through an assisted, interactive process — interviews the user (plan mode + structured questions), drafts a numbered-section spec with EARS acceptance criteria, derives an epic/task backlog from it, review-gates the result, and registers it in project.yaml. Use for 'write/crea

• create-inbound

  Captures an ad-hoc idea, bug, or request onto the board — search-first, so it checks for duplicates before creating anything new. Use for the /create-inbound command with a description, filing a new task without leaving the session, or "add this to the backlog".

• dev-up

  Brings up the project's local dev stack (commands.devUp in .neural-network/project.yaml) so you can QA a change, validate it against acceptance criteria, or debug the running system. Use for 'run the stack', 'start the dev environment', or QA of a merged task — not needed for unit/integration TDD.

• feedback

  Records structured agent feedback about the WORKFLOW itself (not the project being built) — what worked, what caused friction, incidents, recommendations — into the loop's feedback feed for later triage. Use at the end of a build-loop iteration when methodology.feedback is enabled.

• find-task

  Runs a ranked search of existing board issues (open and closed) by title/body similarity, so you can see if something's already been filed. Use for the /find-task command with a search query, checking whether work already exists before filing something new, or "has this been reported before".

• gate

  Runs the project's quality gate via gate.sh — the pass a hook requires before any task can move to In review, and the source of truth for whether the build is green. Use for 'run the gate', 'is it green', and always before review or merge — green is mandatory and enforced.

• handoff

  Writes a handoff document capturing a board snapshot, the work done this session, running state, how to resume, and any gaps or blockers, so the next person or agent can pick up cleanly. Use at a loop checkpoint, end of a working session, 'write a handoff', or when pausing the build loop.

• implement-task

  Implements ONE board task end to end — checks for a design-doc guard, briefs a dev subagent with what/how/why under strict TDD, verifies tests-first plus invariants and spec deltas, runs a two-pass review (spec compliance, then code quality), and drives the board throughout. Use when a specific issu

• knowledge-base-seed

  Explores the current project (specs, backlogs, design docs, applied spec-deltas, READMEs/AGENTS.md/CLAUDE.md, script/source layout, git history, board epics from config) and seeds or updates a knowledge-graph identity brain (.neural-network/identities/knowledge/brain/) via kb-seed.sh, reusing brain.

• migrate-version

  Migrates a consumer repo's spec-workflow files (project.yaml, specs, knowledge bases, feeds) from one schemaVersion to the current one, chapter by chapter — detects the repo's version (missing or integer schemaVersion reads as 1.0.0), then applies each registry chapter in order, verifying after ever

• neural-network-projects-list

  List local repositories anchored with .neural-network and summarize their config, methodology toggles, configured boards and validation errors. Use for a read-only project inventory or to see which local projects participate in spec-workflow.

• neural-view

  Starts, stops, or reports status for the live JARVIS-style visualization of the identity brains — notes shown as neurons, links as synapses, recalls lighting up in real time — plus a project-overview HUD (hover inspection, per-repo board state, best-effort live sessions). Use with 'start', 'stop', o

• next-task

  Picks the next task from the board, honoring priority order, epic sequencing, dependency guards, and the WIP limit, and reads the issue's human comments before committing to it. Use at the start of each build iteration or for 'what should I work on next'.

• pr-review-model

  Shows or changes which models the autonomous PR reviewer is allowed to run on (delegation.identities.reviewer.models), used when auto-merge is on. Use when the user wants a stronger/cheaper/larger-context PR reviewer or asks which model reviews PRs. With no argument, ask the user to pick from option

• queue

  Lists the tasks the autonomous build loop will pick up next, priority-first, presented in human-readable form so you can see what's coming. Use for 'what's next', 'what will build-next pick up', 'show the task queue', or 'what's upcoming'.

• refine-task-ui

  Refines an existing board task's UI through the iterative-UI decision hub, then captures the finalized design as real screenshots and folds them — plus the resolved acceptance criteria — back into the task's issue body. Use for the /refine-task-ui command with a task id — when a task's UI direction

• remote-compute-control

  Inspect installed remote capabilities, read job status or bounded logs, or cancel an existing job through assistant dedicated restricted SSH keys. Use for approved WSL targets configured in this installed skill; use the human remote-compute plugin for registration, trusted recipe dispatch, and admin

• retrospective

  Runs the orchestrator's retro protocol on demand — dedupes and routes any pending feedback, then mints, prunes, or graduates brain notes from it — independent of a build-next PR-close. Use when the human explicitly asks to run a retrospective or "update the brains from feedback", or when the feedbac

• seed-board

  Seeds the GitHub Project board from a spec's backlog — complexity-scores every task (splitting any 8+ before it enters the board), then idempotently creates issues and board items with status, priority, and estimate already set. Use after setup-project, or when a new spec or new tasks are added to t

• setup-assistant

  Scaffolds a bare-brain assistant repo — marker file, project.yaml assistant section, brain directories, persona AGENTS.md, gitignores — interviews the human to compose a real persona, and edits settings (provider/model/capabilities/persona/machine-local default). Use for '/setup-assistant', 'turn th

• setup-project

  Bootstraps a repository for the spec-workflow — creates the GitHub Project board, auto-fills .neural-network/project.yaml ids via init-config.sh, validates the config, and sets up local-state gitignores, so the repo is ready to run the workflow. Use for 'set up this repo', 'adopt the workflow', 'ini

• sync-project-configs

  Discovers every anchored repo (marked with .neural-network) under a scan base and brings its .neural-network/project.yaml up to this plugin's current config surface, via versioned sync rules. Dry-run by default, so nothing changes until you say so. Use when the plugin's config surface has evolved (n

• task-state

  Keeps the local record of which tasks agents are working on and their TODOs, shared by every agent and worktree of a repo and shown in the herdr tasks panel. Use when picking up, progressing, handing over or finishing a task, or when asked what agents are working on right now. It never reads or writ

• ui-mode

  Checks the status of, turns on, or turns off Iterative UI mode, which delegates UI decisions to the human via the decision hub instead of the agent deciding alone. Use with 'status' (default), 'on', or 'off' — when the user asks whether the mode is active, wants UI questions to stop (going AFK), or

• ui-options

  Runs Iterative UI mode's decision step — presents 2-4 concrete UI design options on the local decision hub as toggle tabs (no scrolling), with a favorite selector, likeable-aspect chips, element-level annotations, a final 'Use this one', and a Review tab summarizing every pick before sending, so the

• whisper-sidecar

  Installs, starts, stops, and health-checks the local whisper.cpp speech-to-text sidecar the assistant's voice input relies on when the whisper STT engine is selected. Use for local, private, offline speech transcription -- fully on-machine, no audio leaves the device.

Hooks (7)

• SessionStart

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/session-start.sh"

• SubagentStart

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/herdr-worktree.sh" subagent-hook

• SubagentStop

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/herdr-worktree.sh" subagent-hook

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard-board-move.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard-brain-access.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard-brief-preflight.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard-brain-access.sh"

Similar plugins